SANDIA REPORT

SAND2002-0835
Unlimited Release
Printed April 2002

Combination of Evidence in Dempster-

Shafer Theory

Karl Sentz and Scott Ferson

Prepared by
Sandia National Laboratories
Albuquerque, New Mexico 87185 and Livermore, California 94550
Sandia is a multiprogram laboratory operated by Sandia Corporation,
a Lockheed Martin Company, for the United States Department of
Energy under Contract DE-AC04-94AL85000.

Approved for public release; further dissemination unlimited.

Issued by Sandia National Laboratories, operated for the United States Department
of Energy by Sandia Corporation.

NOTICE: This report was prepared as an account of work sponsored by an agency

of the United States Government. Neither the United States Government, nor any
agency thereof, nor any of their employees, nor any of their contractors,
subcontractors, or their employees, make any warranty, express or implied, or
assume any legal liability or responsibility for the accuracy, completeness, or
usefulness of any information, apparatus, product, or process disclosed, or represent
that its use would not infringe privately owned rights. Reference herein to any
specific commercial product, process, or service by trade name, trademark,
manufacturer, or otherwise, does not necessarily constitute or imply its endorsement,
recommendation, or favoring by the United States Government, any agency thereof,
or any of their contractors or subcontractors. The views and opinions expressed
herein do not necessarily state or reflect those of the United States Government, any
agency thereof, or any of their contractors.

Printed in the United States of America. This report has been reproduced directly
from the best available copy.

Available to DOE and DOE contractors from

U.S. Department of Energy
Office of Scientific and Technical Information
P.O. Box 62
Oak Ridge, TN 37831

Telephone: (865)576-8401
Facsimile: (865)576-5728
E-Mail: reports@adonis.osti.gov
Online ordering: http://www.doe.gov/bridge

Available to the public from

U.S. Department of Commerce
National Technical Information Service
5285 Port Royal Rd
Springfield, VA 22161

Telephone: (800)553-6847
Facsimile: (703)605-6900
E-Mail: orders@ntis.fedworld.gov
Online order: http://www.ntis.gov/ordering.htm
 SAND 2002-0835
Unlimited Release
Printed April 2002

Combination of Evidence in Dempster-Shafer Theory

Kari Sentz
Ph.D. Student
Systems Science and Industrial Engineering Department
Thomas J. Watson School of Engineering and Applied Science
Binghamton University
P.O. Box 6000
Binghamton, NY 13902-6000

Scott Ferson
Applied Biomathematics
100 North Country Road
Setauket, NY 11733

Abstract
Dempster-Shafer theory offers an alternative to traditional probabilistic theory for the
mathematical representation of uncertainty. The significant innovation of this framework
is that it allows for the allocation of a probability mass to sets or intervals. Dempster-
Shafer theory does not require an assumption regarding the probability of the individual
constituents of the set or interval. This is a potentially valuable tool for the evaluation of
risk and reliability in engineering applications when it is not possible to obtain a precise
measurement from experiments, or when knowledge is obtained from expert elicitation.
An important aspect of this theory is the combination of evidence obtained from multiple
sources and the modeling of conflict between them. This report surveys a number of
possible combination rules for Dempster-Shafer structures and provides examples of the
implementation of these rules for discrete and interval-valued data.

3
 ACKNOWLEDGEMENTS

The authors wish to thank Bill Oberkampf, Jon Helton, and Marty Pilch of Sandia
National Laboratories for their many critical efforts in support of this project and the
development of this report in particular. In addition, the initiative of Cliff Joslyn to
organize the workshop on new methods in uncertainty quantification at Los Alamos
National Laboratories (February, 2002) was extremely helpful to the final draft of this
paper. Finally, we would like to thank George Klir of Binghamton University for his
encouragement over the years.

4
 TABLE OF CONTENTS

ABSTRACT.........................................................................................................................3
ACKNOWLEDGEMENTS.................................................................................................4
TABLE OF CONTENTS.....................................................................................................5
LIST OF FIGURES .............................................................................................................6
LIST OF TABLES...............................................................................................................7
1.1: INTRODUCTION ...................................................................................................8
1.2: TYPES OF EVIDENCE....................................................................................10
2.1: DEMPSTER-SHAFER THEORY.........................................................................13
2.2: RULES FOR THE COMBINATION OF EVIDENCE.....................................15
2.2.1: THE DEMPSTER RULE OF COMBINATION.......................................16
2.2.2: DISCOUNT+COMBINE METHOD ........................................................17
2.2.3: YAGER’S MODIFIED DEMPSTER’S RULE.........................................18
2.2.4: INAGAKI’S UNIFIED COMBINATION RULE.....................................20
2.2.5: ZHANG’S CENTER COMBINATION RULE ........................................23
2.2.6: DUBOIS AND PRADE’S DISJUNCTIVE CONSENSUS RULE...........24
2.2.7: MIXING OR AVERAGING .....................................................................25
2.2.8: CONVOLUTIVE X-AVERAGING..........................................................26
2.2.9: OTHER RULES OF COMBINATION.....................................................26
3: DEMONSTRATION OF COMBINATION RULES................................................27
3.1: Data given by discrete values ............................................................................27
3.1.1: Dempster’s Rule ........................................................................................27
3.1.2: Yager’s Rule ..............................................................................................29
3.1.3: Inagaki’s Rule............................................................................................29
3.1.4: Zhang’s Rule..............................................................................................30
3.1.5: Mixing........................................................................................................30
3.1.6: Dubois and Prade’s Disjunctive Consensus Pooling .................................31
3.2: Data given by intervals ......................................................................................31
3.2.1: Dempster’s Rule ..............................................................................................34
3.2.2: Yager’s Rule ....................................................................................................35
3.2.3: Inagaki’s Rule..................................................................................................36
3.2.4: Zhang’s Rule....................................................................................................36
3.2.5: Mixing..............................................................................................................39
3.2.6: Convolutive x-Averaging ................................................................................40
3.2.7 Dubois and Prade’s Disjunctive Consensus......................................................43
3.2.8: Summary of Examples.....................................................................................45
4: CONCLUSIONS .......................................................................................................46

REFERENCES ..................................................................................................................50

APPENDIX A................................................................................................................. A-1

5
 LIST OF FIGURES

Figure 1: Consonant evidence obtained from multiple sources.........................................11

Figure 2: Consistent evidence obtained from multiple sensors.........................................11
Figure 3: Arbitrary evidence obtained from multiple sensors...........................................12
Figure 4: Disjoint evidence obtained from multiple sensors.............................................12
Figure 5: The Possible Values of k in Inagaki’s Unified Combination Rule ....................22
Figure 6: The value of m(B) as a function of k in Inagaki’s rule......................................30
Figure 7: The gcdf of A .....................................................................................................32
Figure 8: The gcdf of B......................................................................................................32
Figure 9: The gcdf of C......................................................................................................33
Figure 10: The gcdf’s of A, B, and C without any combination operation.......................33
Figure 11: The gcdf of the combination of A and B using Dempster’s rule .....................34
Figure 12: The combination of A and B using Yager’s rule .............................................35
Figure 13: The gcdf of the combination of A and C using Yager’s rule ...........................35
Figure 14: The Inagaki combination of A and B for k=0 ..................................................36
Figure 15: The Inagaki combination of A and B where k = 1 ...........................................36
Figure 16: The Zhang combination of A and B.................................................................38
Figure 17: The mixture of A and B....................................................................................39
Figure 18: The mixture of A and C....................................................................................40
Figure 19: The gcdf of the combination of A and B using convolutive x-averaging ........41
Figure 20: The Comparison of Combinations of A and B with Dempster’s rule and
Convolutive X-Averaging..........................................................................................41
Figure 21: The gcdf of the Combination of A and C using Convolutive x-Averaging .....42
Figure 22: Comparison of Yager’s rule and Convolutive x-averaging for A and C .........43
Figure 23: The Disjunctive Consensus Pooling of A and B ..............................................44
Figure 24: The gcdf for the Disjunctive Consensus Pooling of A and C ..........................44
Figure 25: Important Issues in the Combination of Evidence ...........................................48

6
 LIST OF TABLES

Table 1: Dempster Combination of Expert 1 and Expert 2 ...............................................28

Table 2: Unions obtained by Disjunctive Consensus Pooling...........................................31
Table 3: The interval-based data for A and the basic probability assignments .................31
Table 4: The interval-based data for B and the basic probability assignments .................32
Table 5: The interval-based data for C and the basic probability assignments .................33
Table 6: Combination of A and B with Dempster’s Rule.................................................34
Table 7: The combination of the marginals with Zhang’s rule..........................................37
Table 8: The length of the intervals and their intersections...............................................37
Table 9: Calculation of the Measure of Intersection..........................................................37
Table 10: The product of m and r(A,B)..............................................................................38
Table 11: The renormalized masses with Zhang’s rule .....................................................38
Table 12: The mixture of A and B.....................................................................................39
Table 13: The mixture of A and C.....................................................................................40
Table 14: The Combination of A and B using Convolutive x-Averaging.........................40
Table 15: The Combination of A and C using Convolutive x-Averaging.........................42
Table 16: The Disjunctive Consensus Pooling of A and B ...............................................43
Table 17: Calculations for the Disjunctive Consensus Pooling of A and C ......................44
Table 18: The Combination of A and B Comparison Table..............................................45
Table 19: The Combination of A and C Comparison Table..............................................46
Table 20: Combination Rules and Their Algebraic Properties..........................................47

7
1.1: INTRODUCTION
Only very recently, the scientific and engineering community has begun to
recognize the utility of defining multiple types of uncertainty. In part the greater depth of
study into the scope of uncertainty is made possible by the significant advancements in
computational power we now enjoy. As systems become computationally better
equipped to handle complex analyses, we encounter the limitations of applying only one
mathematical framework (traditional probability theory) used to represent the full scope
of uncertainty. The dual nature of uncertainty is described with the following definitions
from [Helton, 1997]:

Aleatory Uncertainty – the type of uncertainty which results from the fact that a
system can behave in random ways
also known as: Stochastic uncertainty, Type A uncertainty, Irreducible
uncertainty, Variability, Objective uncertainty

Epistemic Uncertainty- the type of uncertainty which results from the lack of
knowledge about a system and is a property of the analysts performing the
analysis.
also known as: Subjective uncertainty, Type B uncertainty, Reducible uncertainty,
State of Knowledge uncertainty, Ignorance

Traditionally, probability theory has been used to characterize both types of

uncertainty. It is well recognized that aleatory uncertainty is best dealt with using the
frequentist approach associated with traditional probability theory. However, the recent
criticisms of the probabilistic characterization of uncertainty claim that traditional
probability theory is not capable of capturing epistemic uncertainty. The application of
traditional probabilistic methods to epistemic or subjective uncertainty is often known as
Bayesian probability. A probabilistic analysis requires that an analyst have information
on the probability of all events. When this is not available, the uniform distribution
function is often used, justified by Laplace’s Principle of Insufficient Reason. [Savage,
1972] This can be interpreted that all simple events for which a probability distribution is
not known in a given sample space are equally likely. Take for an example a system
failure where there are three possible components that could have caused this type of
failure. An expert in the reliability of one component assigns a probability of failure of
that component with 0.3 (Component A). The expert knows nothing about the other two
potential sources of failure (Components B and C). A traditional probabilistic analysis
following the Principle of Insufficient Reason, could assign a probability of failure of
0.35 to each of the two remaining components (B and C). This would be a very precise
statement about the probability of failure of these two components in the face of complete
ignorance regarding these components on the part of the expert.
An additional assumption in classical probability is entailed by the axiom of
additivity where all probabilities that satisfy specific properties must add to 1. This
forces the conclusion that knowledge of an event necessarily entails knowledge of the
complement of an event, i.e., knowledge of the probability of the likelihood of the
occurrence of an event can be translated into the knowledge of the likelihood of that
event not occurring. If an expert believes that a system may fail due to a particular

8
component with a likelihood of 0.3, does that necessarily mean that the expert believes
that the system will not fail due to that component of 0.7? This articulates the challenge
of modeling any uncertainty associated with an expert’s subjective belief. Though the
assumptions of additivity and the Principle of Insufficient Reason may be appropriate
when modeling the random events associated with aleatoric uncertainty, these constraints
are questionable when applied to an issue of knowledge or belief.
As a consequence of these concerns, applied mathematicians have investigated
many more general representation of uncertainty to cope with particular situations
involving epistemic uncertainty. Examples of these types of situations include:

1. When there is little information on which to evaluate a probability or

2. When that information is nonspecific, ambiguous, or conflicting.

Analysis of these situations can be required, for an example in risk assessment, though
probability theory lacks the ability to handle such information. Where it is not possible to
characterize uncertainty with a precise measure such as a precise probability, it is
reasonable to consider a measure of probability as an interval or a set.
This characterization of a measure of probability as an interval or set has three
important implications:

1. It is not necessary to elicit a precise measurement from an expert or an

experiment if it is not realistic or feasible to do so.
2. The Principle of Insufficient Reason is not imposed. Statements can be made
about the likelihood of multiple events together without having to resort to
assumptions about the probabilities of the individual events under ignorance.
3. The axiom of additivity is not imposed. The measures do not have to add to 1.
When they do, it corresponds to a traditional probabilistic representation.
When the sum is less than 1, called the subadditive case, this implies an
incompatibility between multiple sources of information, e.g. multiple sensors
providing conflicting information. When the sum is greater than 1, the
superadditive case, this implies a cooperative effect between multiple sources
of information, e.g. multiple sensors providing the same information.

Because there is more than one kind of uncertainty and probability theory may not
apply to every situation involving uncertainty, many theories of generalized uncertainty-
based information have been developed. Currently, this discipline area is known as
monotone measure theory or nonadditive measure theory but in older publications it is
referred to as fuzzy measure theory. This latter designation is a misnomer as the majority
of frameworks subsumed under this term are not fuzzy in the traditional use of the term
as introduced by Zadeh. There are three major frameworks from which the problem of
interval-based representation of uncertainty has been approached: imprecise probabilities
(initial work by Walley, Fine; Kuznetsov); possibility theory (Zadeh; Dubois and Prade;
Yager); and the Dempster-Shafer theory of evidence. (Dempster; Shafer; Yager; Smets).
This situation of multiple frameworks to characterize uncertainty poses an
obvious problem to the analyst faced with epistemic uncertainty, namely, which method
should be applied to a particular situation. While this is still a research question, this

9
decision is simplified somewhat by the level of development of the theories and their use
in practical applications. This study uses Dempster-Shafer Theory as the framework for
representing uncertainty and investigates the issue of combination of evidence in this
theory. The motivation for selecting Dempster-Shafer theory can be characterized by the
following reasons:

1. The relatively high degree of theoretical development among the non-

traditional theories for characterizing uncertainty.
2. The relation of Dempster-Shafer theory to traditional probability theory and
set theory.
3. The large number of examples of applications of Dempster-Shafer theory in
engineering in the past ten years.
4. The versatility of the Dempster-Shafer theory to represent and combine
different types of evidence obtained from multiple sources.

1.2: TYPES OF EVIDENCE

There are two critical and related issues concerning the combination of evidence
obtained from multiple sources: one is the type of evidence involved and the other is how
to handle conflicting evidence. We consider four types of evidence from multiple
sources that impact the choice of how information is to be combined: consonant
evidence, consistent evidence, arbitrary evidence, and disjoint evidence:

Consonant evidence can be represented as a nested structure of subsets where the

elements of the smallest set are included in the next larger set… all of whose elements are
included in the next larger set and so on. This can correspond to the situation where
information is obtained over time that increasingly narrows or refines the size of the
evidentiary set. Take a simple example from target identification. Suppose there are five
sensors with varying degrees of resolution: Sensor 1; Sensor 2; Sensor 3; Sensor 4;
Sensor 5.

Sensor 1 detects a target in vicinity A.

Sensor 2 detects two targets: one in vicinity A and one in vicinity B.
Sensor 3 detects three targets: one in vicinity A, one in vicinity B, one in vicinity
C.
Sensor 4 detects four targets: one in vicinity A, one in vicinity B, one in vicinity
C, one in vicinity D.
Sensor 5 detects five targets: one in vicinity A, one in vicinity B, one in vicinity
C, one in vicinity, one in vicinity E.

10
 A A
B
B
E
C
C D
E
D

Figure 1: Consonant evidence obtained from multiple sources

Consistent evidence means that there is at least one element that is common to all
subsets. From our target identification, this could look like:

Sensor 1 detects a target in vicinity A.

Sensor 2 detects two targets: one in vicinity A and one in vicinity B.
Sensor 3 detects two targets: one in vicinity A, one in vicinity C.
Sensor 4 detects three targets: one in vicinity A, one in vicinity B, one in vicinity
D.
Sensor 5 detects four targets: one in vicinity A, one in vicinity B, one in vicinity
C, one in vicinity E.

A
A B
E B
C
C D
E
D

Figure 2: Consistent evidence obtained from multiple sensors

Arbitrary evidence corresponds to the situation where there is no element common to
all subsets, though some subsets may have elements in common. One possible
configuration in our target identification example:

Sensor 1 detects a target in vicinity A.

Sensor 2 detects two targets: one in vicinity A and one in vicinity B.

11
 Sensor 3 detects two targets: one in vicinity A, one in vicinity C.
Sensor 4 detects two targets: one in vicinity C, one in vicinity D.
Sensor 5 detects two targets: one in vicinity C, one in vicinity E.

A A
B
B
E
C
C
D
D E

Figure 3: Arbitrary evidence obtained from multiple sensors

Disjoint evidence implies that any two subsets have no elements in common with any
other subset.
Sensor 1 detects a target in vicinity A.
Sensor 2 detects a target in vicinity B.
Sensor 3 detects a target in vicinity C.
Sensor 4 detects a target in vicinity D.
Sensor 5 detects a target in vicinity E.

A
B
A B
E
C
C D
E
D

Figure 4: Disjoint evidence obtained from multiple sensors

Each of these possible configurations of evidence from multiple sources has
different implications on the level of conflict associated with the situation. Clearly in the
case of disjoint evidence, all of the sources supply conflicting evidence. With arbitrary
evidence, there is some agreement between some sources but there is no consensus
among sources on any one element. Consistent evidence implies an agreement on at least
one evidential set or element. Consonant evidence represents the situation where each set
is supported by the next larger set and implies an agreement on the smallest evidential
set; however, there is conflict between the additional evidence that the larger set
represents in relation to the smaller set. Traditional probability theory cannot handle

12
consonant, consistent, or arbitrary evidence without resorting to further assumptions of
the probability distributions within a set, nor can probability theory express the level of
conflict between these evidential sets. Dempster-Shafer theory is a framework that can
handle these various evidentiary types by combining a notion of probability with the
traditional conception of sets. In addition, in Dempster Shafer theory, there are many
ways which conflict can be incorporated when combining multiple sources of
information.

2.1: DEMPSTER-SHAFER THEORY

Dempster-Shafer Theory (DST) is a mathematical theory of evidence. The

seminal work on the subject is [Shafer, 1976], which is an expansion of [Dempster,
1967]. In a finite discrete space, Dempster-Shafer theory can be interpreted as a
generalization of probability theory where probabilities are assigned to sets as opposed to
mutually exclusive singletons. In traditional probability theory, evidence is associated
with only one possible event. In DST, evidence can be associated with multiple possible
events, e.g., sets of events. As a result, evidence in DST can be meaningful at a higher
level of abstraction without having to resort to assumptions about the events within the
evidential set. Where the evidence is sufficient enough to permit the assignment of
probabilities to single events, the Dempster-Shafer model collapses to the traditional
probabilistic formulation. One of the most important features of Dempster-Shafer theory
is that the model is designed to cope with varying levels of precision regarding the
information and no further assumptions are needed to represent the information. It also
allows for the direct representation of uncertainty of system responses where an
imprecise input can be characterized by a set or an interval and the resulting output is a
set or an interval.
There are three important functions in Dempster-Shafer theory: the basic
probability assignment function (bpa or m), the Belief function (Bel), and the Plausibility
function (Pl).
The basic probability assignment (bpa) is a primitive of evidence theory.
Generally speaking, the term “basic probability assignment” does not refer to probability
in the classical sense. The bpa, represented by m, defines a mapping of the power set to
the interval between 0 and 1, where the bpa of the null set is 0 and the summation of the
bpa’s of all the subsets of the power set is 1. The value of the bpa for a given set A
(represented as m(A)), expresses the proportion of all relevant and available evidence that
supports the claim that a particular element of X (the universal set) belongs to the set A
but to no particular subset of A [Klir, 1998]. The value of m(A) pertains only to the set A
and makes no additional claims about any subsets of A. Any further evidence on the
subsets of A would be represented by another bpa, i.e. B ⊂ A, m(B) would the bpa for the
subset B. Formally, this description of m can be represented with the following three
equations:
m: P (X)→[0,1] (1)

m(∅) = 0 (2)

13
 ∑ m(A) = 1 (3)
A∈P (X )

where P (X) represents the power set of X, ∅ is the null set, and A is a set in the power set
(A∈ P (X)). [Klir, 1998]
Some researchers have found it useful to interpret the basic probability
assignment as a classical probability, such as [Chokr and Kreinovich, 1994], and the
framework of Dempster-Shafer theory can support this interpretation. The theoretical
implications of this interpretation are well developed in [Kramosil, 2001]. This is a very
important and useful interpretation of Dempster-Shafer theory but it does not demonstrate
the full scope of the representational power of the basic probability assignment. As such,
the bpa cannot be equated with a classical probability in general.
From the basic probability assignment, the upper and lower bounds of an interval
can be defined. This interval contains the precise probability of a set of interest (in the
classical sense) and is bounded by two nonadditive continuous measures called Belief
and Plausibility. The lower bound Belief for a set A is defined as the sum of all the basic
probability assignments of the proper subsets (B) of the set of interest (A) (B ⊆ A). The
upper bound, Plausibility, is the sum of all the basic probability assignments of the sets
(B) that intersect the set of interest (A) (B ∩ A ≠ ∅). Formally, for all sets A that are
elements of the power set (A∈ P (X)), [Klir, 1998]

Bel ( A) = ∑ m( B)
B| B ⊆ A
(4)

(5)
Pl ( A ) = ∑ m(B )
B |B ∩ A ≠ ∅

The two measures, Belief and Plausibility are nonadditive. This can be interpreted as is
not required for the sum of all the Belief measures to be 1 and similarly for the sum of the
Plausibility measures.
It is possible to obtain the basic probability assignment from the Belief measure
with the following inverse function:
∑ (−1)
A −B
m( A) = Bel ( B) (6)
B| B ⊆ A

where |A-B| is the difference of the cardinality of the two sets.

In addition to deriving these measures from the basic probability assignment

(m), these two measures can be derived from each other. For example, Plausibility can
be derived from Belief in the following way:

Pl ( A) = 1 − Bel ( A) (7)

where A is the classical complement of A. This definition of Plausibility in terms of

Belief comes from the fact that all basic assignments must sum to 1.

14
 Bel(A) = ∑ m(B) = ∑ m(B) (8)
B B⊆ A B B∩ A=∅

∑ m(B) = 1 − ∑ m(B)
B B ∩A≠ ∅ B B∩ A=∅
(9)

From the definitions of Belief and Plausibility, it follows that Pl ( A) = 1 − Bel ( A) . As a

consequence of Equations 6 and 7, given any one of these measures (m(A), Bel(A), Pl(A))
it is possible to derive the values of the other two measures.
The precise probability of an event (in the classical sense) lies within the lower
and upper bounds of Belief and Plausibility, respectively.

Bel(A) = P(A) = Pl(A) (10)

The probability is uniquely determined if Bel (A) = Pl(A). In this case, which
corresponds to classical probability, all the probabilities, P(A) are uniquely determined
for all subsets A of the universal set X [Yager, 1987, p.97]. Otherwise, Bel (A) and Pl(A)
may be viewed as lower and upper bounds on probabilities, respectively, where the actual
probability is contained in the interval described by the bounds. Upper and lower
probabilities derived by the other frameworks in generalized information theory can not
be directly interpreted as Belief and Plausibility functions. [Dubois and Prade, 1992,
p.216]

2.2: RULES FOR THE COMBINATION OF EVIDENCE

The purpose of aggregation of information is to meaningfully summarize and

simplify a corpus of data whether the data is coming from a single source or multiple
sources. Familiar examples of aggregation techniques include arithmetic averages,
geometric averages, harmonic averages, maximum values, and minimum values [Ayuub,
2001]. Combination rules are the special types of aggregation methods for data obtained
from multiple sources. These multiple sources provide different assessments for the same
frame of discernment and Dempster-Shafer theory is based on the assumption that these
sources are independent. The requirement for establishing the independence of sources is
an important philosophical question.
From a set theoretic standpoint, these rules can potentially occupy a continuum
between conjunction (AND-based on set intersection) and disjunction (OR-based on set
union) [Dubois and Prade, 1992]. In the situation where all sources are considered
reliable, a conjunctive operation is appropriate (A and B and C…). In the case where
there is one reliable source among many, we can justify the use of a disjunctive
combination operation (A or B or C…). However, many combination operations lie
between these two extremes (A and B or C, A and C or B, etc.). Dubois and Prade
[Dubois, Prade, 1992] describe these three types of combinations as conjunctive pooling
(A∩B, if A∩B≠∅), disjunctive pooling (A∪B), and tradeoff (There are many ways a
tradeoff between A∩B and A∪B can be achieved).
There are multiple operators available in each category of pooling by which a
corpus of data can be combined. One means of comparison of combination rules is by

15
comparing the algebraic properties they satisfy. With the tradeoff type of combination
operations, less information is assumed than in a Bayesian approach and the precision of
the result may suffer as a consequence. On the other hand, a precise answer obtained via
the Bayesian approach does not express any uncertainty associated with it and may have
hidden assumptions of additivity or Principle of Insufficient Reason. [Dubois and Prade,
1992]
In keeping with this general notion of a continuum of combination operations,
there are multiple possible ways in which evidence can be combined in Dempster-Shafer
theory. The original combination rule of multiple basic probability assignments known
as the Dempster rule is a generalization of Bayes’ rule. [Dempster, 1967] This rule
strongly emphasizes the agreement between multiple sources and ignores all the
conflicting evidence through a normalization factor. This can be considered a strict
AND-operation. The use of the Dempster rule has come under serious criticism when
significant conflict in the information is encountered. [Zadeh, 1986; Yager, 1987]
Consequently, other researchers have developed modified Dempster rules that attempt to
represent the degree of conflict in the final result. This issue of conflict and the
allocation of the bpa mass associated with it is the critical distinction between all of the
Dempster-type rules. To employ any of these combination rules in an application, it is
essential to understand how conflict should be treated in that particular application
context.
In addition to the Dempster rule of combination, we will discuss four modified
Dempster rules: Yager’s rule; Inagaki’s unified combination rule; Zhang’s center
combination rule; and Dubois and Prade’s disjunctive pooling rule. Three types of
averages will be considered: discount and combine; convolutive averaging; and mixing.
All of the combination rules will be considered relative to four algebraic properties:
commutativity, A * B = B * A; idempotence, A * A = A; continuity, A * B ≈ A′ * B,
where A′≈A (A′ is very close to A); and associativity, A * (B * C) = (A * B) * C; where
* denotes the combination operation. The motivation for these properties is discussed at
length in [Ferson and Kreinovich, 2002].

2.2.1: THE DEMPSTER RULE OF COMBINATION

The Dempster rule of combination is critical to the original conception of

Dempster-Shafer theory. The measures of Belief and Plausibility are derived from the
combined basic assignments. Dempster’s rule combines multiple belief functions
through their basic probability assignments (m). These belief functions are defined on
the same frame of discernment, but are based on independent arguments or bodies of
evidence. The issue of independence is a critical factor when combining evidence and is
an important research subject in Dempster-Shafer theory. The Dempster rule of
combination is purely a conjunctive operation (AND). The combination rule results in a
belief function based on conjunctive pooled evidence [Shafer, 1986, p.132].
Specifically, the combination (called the joint m12) is calculated from the
aggregation of two bpa’s m1 and m2 in the following manner:

∑ m (B)m (C)
B∩C = A
1 2

m12( A) = when A≠∅ (11)

1− K

16
 m 12(∅) = 0 (12)

where K = ∑ m (B)m (C)

B∩C =∅
1 2 (13)

K represents basic probability mass associated with conflict. This is determined by the
summing the products of the bpa’s of all sets where the intersection is null. This rule is
commutative, associative, but not idempotent or continuous.
The denominator in Dempster’s rule, 1-K, is a normalization factor. This has the
effect of completely ignoring conflict and attributing any probability mass associated with
conflict to the null set [Yager, 1987]. Consequently, this operation will yield
counterintuitive results in the face of significant conflict in certain contexts. The problem
with conflicting evidence and Dempster’s rule was originally pointed out by Lotfi Zadeh
in his review of Shafer’s book, A Mathematical Theory of Evidence [Zadeh, 1984].
Zadeh provides a compelling example of erroneous results. Suppose that a patient is seen
by two physicians regarding the patient’s neurological symptoms. The first doctor
believes that the patient has either meningitis with a probability of 0.99 or a brain tumor,
with a probability of 0.01. The second physician believes the patient actually suffers
from a concussion with a probability of 0.99 but admits the possibility of a brain tumor
with a probability of 0.01. Using the values to calculate the m (brain tumor) with
Dempster’s rule, we find that m(brain tumor) = Bel (brain tumor) = 1. Clearly, this rule
of combination yields a result that implies complete support for a diagnosis that both
physicians considered to be very unlikely. [Zadeh, 1984, p.82]
In light of this simple but dramatic example of the counterintuitive results of
normalization factor in Dempster’s rule, a number of methods and combination
operations that have been developed to address this problem posed by strongly
conflicting evidence. We will discuss many of these alternatives in the following
sections as well as the importance of conflict and context in the rule selection. We will
find that in addition to the level or degree of conflict is important in determining the
propriety of using Dempster's rule, the relevance of conflict also plays a critical role.

2.2.2: DISCOUNT+COMBINE METHOD

This tradeoff method was initially discussed in [Shafer, 1976] and deals with
conflict just in the manner that the name implies. Specifically, when an analyst is faced
with conflicting evidence, he/she can discount the sources first, and then combine the
resulting functions with Dempster’s rule (or an alternative rule) using a discounting
function. This discounting function must account for the absolute reliability of the
sources. Absolute reliability implies that the analyst is qualified to make distinctions
between the reliability of experts, sensors, or other sources of information and can
express this distinction between sources mathematically. [Dubois and Prade, 1992]
Shafer applies the discounting function to each specified Belief. Let 1-αi be the
degree of reliability attributable to a particular belief function, A (Shafer calls this a
degree of trust), where 0 ≤ α i ≤1 and i is an index used to specify the particular

17
discounting function associated with a particular belief measure. Belαi (A) then represents
the discounted belief function defined by:
αi
Bel ( A) = (1 − α i ) Bel ( A) (14)

Shafer then averages all the belief functions associated with set A (Belαi 1(A),
Belαi 2(A)…. Belαi n(A)) to obtain an average of n Bel, denoted by Bel .
1
Bel ( A) = ( Bel α 1 ( A) + ... + Bel α n ( A)) (15)
n

for all subsets A of the universal set X.

Consequently, the discount and combine method uses an averaging function as the
method of combination. This is to be used when all the belief functions to be combined
are highly conflicting and the discounting rate is not too small. This can also be used to
eliminate the influence of any strongly conflicting single belief function provided that the
remaining belief functions do not conflict too much with each other and the discount rate
is not too small or too large. Alternatively, for this case one could also eliminate the
strongly conflicting belief altogether if that is reasonable. [Shafer, 1976]

2.2.3: YAGER’S MODIFIED DEMPSTER’S RULE

The most prominent of the alternative combination rules is a class of unbiased

operators developed by Ron Yager. [Yager, 1987a] Yager points out that an important
feature of combination rules is the ability to update an already combined structure when
new information becomes available. This is frequently referred to as updating and the
algebraic property that facilitates this is associativity. Dempster’s rule is an example of
an associative combination operation and the order of the information does not impact the
resulting fused structure. [Yager, 1987b]
Yager points out that in many cases a non-associative operator is necessary for
combination. A familiar example of this is the arithmetic average. The arithmetic
average is not itself associative, i.e., one cannot update the information by averaging an
average of a given body of data and a new data point to yield a meaningful result.
However, the arithmetic average can be updated by adding the new data point to the sum
of the pre-existing data points and dividing by the total number of data points. This is the
concept of a quasi-associative operator that Yager introduced in [Yager, 1987b]. Quasi-
associativity means that the operator can be broken down into associative suboperations.
Through the notion of quasi-associative operator, Yager develops a general framework to
look at combination rules where associative operators are a proper subset.
To address the issue of conflict, Yager starts with an important distinction
between the basic probability mass assignment (m) and what he refers to as the ground
probability mass assignment (designated by q). The major differences between the basic
probability assignment and the ground probability assignment are in the normalization
factor and the mass attributed to the universal set. The combined ground probability
assignment is defined in equation 16.

18
 q( A) = ∑m (B)m (C)
B∩C=A
1 2 (16)

where A is the intersection of subsets B and C (both in the power set P (X)), and q(A)
denotes the ground probability assignment associated with A. Note that there is no
normalization factor. This rule is known as Yager’s combination rule or sometimes the
Modified Dempster’s Rule.
Though the Yager rule of combination is not associative, the combined structure
q(A) can be used to include any number of pieces of evidence. Assume m1, m2,…mn are
the basic probability assignments for n belief structures. Let Fi represent the set of focal
elements associated with the ith belief structure (mi) which are subsets of the universal set
X. Ai represents an element of the focal set. Then the combination of n basic probability
assignment structures is defined by [Yager, 1987a]:
(17)
q(A) = ∑ m 1 ( A 1 ) m 2 ( A 2 )... m n ( A n )
∩ n A = A
i=1 i

Through the quasiassociativity that Yager describes, the combined structure q(A) can be
updated based on new evidence. This is performed by combining the ground probability
assignment associated with the new evidence and the ground probability assignment of
the already existing combination through the above formulas (Equation 16) and then
converting the ground probability assignments to basic probability assignments described
below. (Equations 19-21)
As previously mentioned, one obvious distinction between combination with the
basic and the ground probability assignment functions is the absence of the normalization
factor (1-K). In Yager’s formulation, he circumvents normalization by allowing the
ground probability mass assignment of the null set to be greater than 0, i.e.

q( ∅ ) ≥ 0 (18)

q(∅) is calculated in exactly in the same manner as Dempster’s K (conflict) in Equation

13. Then Yager adds the value of the conflict represented by q(∅) to the ground
probability assignment of the universal set, q(X), to yield the conversion of the ground
probabilities to the basic probability assignment of the universal set mY(X):

m (X ) = q( X ) + q(∅)
Y
(19)

Consequently, instead of normalizing out the conflict, as we find in the case of the
Dempster rule, Yager ultimately attributes conflict to the universal set X through the
conversion of the ground probability assignment to the basic probability assignments.
The interpretation of the mass of the universal set (X) is the degree of ignorance.
Dempster’s rule has the effect of changing the evidence through the normalization and
the allocation of conflicting mass to the null set. Yager’s rule can be considered as an
epistemologically honest interpretation of the evidence as it does not change the evidence
by normalizing out the conflict. In Yager’s rule, the mass associated with conflict is
attributed to the universal set and thus enlarges this degree of ignorance. [Yager, 1987a]

19
Upon inspection of the two combination formulas it is clear that Yager’s rule of
combination yields the same result as Dempster’s rule when conflict is equal to zero, (K
= 0 or q(∅) = 0). [Yager, 1987a] The basic algebraic properties that this rule satisfies is
commutativity and quasiassociativity, but not idempotence or continuity.
The ground probability assignment functions (q) for the null set, ∅, and an
arbitrary set A, are converted to the basic probability assignment function associated with
this Yager’s rule (mY) by [Yager 1987a]:

m (∅ ) = 0
Y
(20)

m ( A ) = q( A )
Y
(21)

The basic probability assignments associated with Yager’s rule (mY) are not the same as
with Dempster’s rule (m). Yager provides the relation between the ground assignments
and Dempster’s rule [Yager 1987a]:

m (∅ ) = 0 (22)
q( X )
m(X) = (23)
1− q(∅)

q(A)
m(A) = (24)
1 − q(∅)

for A ≠ ∅, X

To summarize, these are the important attributes of Yager’s rule of combination:

1. The introduction of the general notion of quasi-associative operators and the

expansion of the theoretical basis for the combination and updating of
evidence where the associative operators are a proper subset of the quasi-
associative operators.
2. The introduction of the ground probability assignment functions (q) and their
relation to the basic probability assignments (mY) associated with Yager’s rule
and the basic probability assignments (m) associated with Dempster’s rule.
3. The rule does not filter or change the evidence through normalization.
4. The allocation of conflict to the universal set (X) instead of to the null set (∅).
Thus mass associated with conflict is interpreted as the degree of ignorance.

2.2.4: INAGAKI’S UNIFIED COMBINATION RULE

This combination rule was introduced by Toshiyuki Inagaki. [Inagaki, 1991]

Inagaki takes advantage of the ground probability assignment function (q) that Yager
defined in [Yager, 1987a] to define a continuous parametrized class of combination
operations which subsumes both Dempster’s rule and Yager’s rule. Specifically, Inagaki
argues that every combination rule can be expressed as:

20
 m (C ) = q ( C ) + f ( C ) q (∅ ) (25)

where C ≠ ∅

(26)
∑ f (C ) = 1
C ⊂ X ,C ≠ ∅

f (C ) ≥ 0 (27)

From Equation 25 the function, f, can be interpreted as a scaling function for q(∅), where
the conflict (represented by the parameter k) is defined by:

f (C )
k= for any C ≠ X, ∅ (28)
q(C)

Inagaki restricts consideration to the class of combination rules that satisfy the
following property:
m(C) q(C)
= (29)
m(D) q( D)

for any nonempty sets C and D which are distinct from X or ∅. By maintaining the ratio
between m and q consistently, this equation implies that there is no “meta-knowledge” of
the credibility or reliability of sources/experts. If an analyst applied a weighting factor to
the evidence based on some extra knowledge about the credibility of the sources, in
general, this would change the ratio and the equality would not hold. As a result of this
restriction and its implication, Inagaki’s rule applies only to the situations where there is
no information regarding the credibility or reliability of the sources. [Inagaki, 1991]
From the general expression (Equation 25) and the restriction (Equation 26) and
the definition of k (Equation 28), Inagaki derives his unified combination rule denoted by
mU.
mU k
( C ) = [1 + kq ( ∅ )] q ( C ), where C ≠ X , ∅ (30)

m k ( X ) = [1 + kq (∅ )] q ( X ) + [1 + kq (∅ ) − k ] q (∅ )
U
(31)

1
0≤k≤ (32)
1− q(∅) − q( X)

The parameter k is used for normalization. The determination of k is an important

step in the implementation of this rule, however, a developed well-justified procedure for
determining k is lacking in the literature reviewed for this report. Tanaka and Klir refer
to the determination of k either through experimental data, simulation, or the expectations
of an expert in the context of a specific application. In addition, they provide an example
for the determination of k and the resulting affect on m for monitoring systems [Tanaka

21
and Klir, 1999]. In [Inagaki, 1991], Inagaki poses the optimization problem for the
selection of k to be an open and critical research question. Despite this, Inagaki discusses
the rules in the context of an application where he demonstrates the values of Belief and
Plausibility as a function of k and the implications on the choice of a safety control
policy.
The value of k directly affects the value of the combined basic probability
assignments and will collapse to either Dempster’s rule or Yager’s rule under certain
circumstances. When k = 0, the unified combination rule coincides with Yager’s rule.
When k = 1 , the rule corresponds to Dempster’s rule. The parameter k gives
1 − q(∅ )
rise to an entire parametrized class of possible combination rules that interpolate or
extrapolate Dempster’s rule. [Inagaki, 1991] This is schematically represented in the
Figure 5 from [Inagaki, 1991]:

Yager’s Dempster’s
rule rule

1 1
0
1 − q(∅) 1 − q(∅) − q( X)

Figure 5: The Possible Values of k in Inagaki’s Unified Combination Rule

The only combination rule of this parametrized class that is associative is the one that
corresponds to Dempster’s rule. Every combination rule represented by the unified
combination rule is commutative though not idempotent or continuous. Inagaki considers
the effect of non-associativity in applications to be an open research question. [Inagaki,
1991]
As is pointed out by Tanaka and Klir [Tanaka and Klir, 1999], the most extreme
rule (referred to as “the extra rule” and denoted by the parameter kext ) availed by this
formulation is when k is equal to the upper bound:

1 − q(X )
m Ukext (C ) = q(C) (33)
1 − q( X ) − q( ∅)

for C ≠ X,

m Uk ext ( X ) = q ( X ) (34)

1 − q( X)
As can be seen in Equation 33, the value of q(C) is scaled by the factor,
1 − q(X) − q(∅)
U
to yield the corresponding basic probability function mkext . The interpretation of the

22
extreme rule of Inagaki’s class is that both conflict (represented by q(∅)) and the degree
of ignorance (represented by the probability mass associated with the universal set, q(X))
are used to scale the resulting combination. This acts as a filter for the evidence.
Inagaki studied the ordering relations of the three rules: Dempster’s rule, Yager’s
rule, and this “extra rule” and the propriety of their application in fault-warning safety
control policy. [Inagaki, 1991] Tanaka and Klir point out that the selection of the
parameter k essentially determines how to cope with conflicting information. Yager’s rule
(k=0) assigns conflict to the universal set and does not change the evidence. Dempster’s
rule (k=1/[1-q(∅]) tremendously filters the evidence by ignoring all conflict. Inagaki’s
extreme rule (k=1/[1-q(∅)-q(X)]) also filters the evidence by scaling both conflict and
ignorance, but the degree of influence of the scaling is determined by the relative values
of q(X) and q(∅). k has the effect of scaling the importance of conflict as it is represented
in the resulting combination. The greater the value of k, the greater the change to the
evidence. As noted earlier, a well-justified procedure for the selection of k is as essential
step toward implementing this rule in an application.
The important contributions of Inagaki’s Unified rule of combination can be
summarized as follows:

1. The use of Yager’s ground functions to develop a parametrized class of

combination rules that subsumes both Dempster’s rule and Yager’s rule.
2. Inagaki compares and orders three combination rules: Dempster’s rule,
Yager’s rule, and the Inagaki extra rule, in terms of the value of m in the
context of an application.

2.2.5: ZHANG’S CENTER COMBINATION RULE

Lianwen Zhang [Zhang, 1994] also provides an alternative combination rule to

Dempster’s rule. In addition, he offers a two frame interpretation of Dempster-Shafer
theory: Suppose there are two frames of discernment, S and T. These could be the
opinions of two experts. Between these frames is a compatibility relation, C, which is a
subset of the Cartesian product S × T. We are concerned with the truth in T but the only
available probability P is about the truth in S. Because of this compatibility relation it
follows that information about S provides some information of T. This information is
summarized as a Belief function for any subset of A of T. The belief function for A can
be written as:

Bel(A) = P{s|s ∈ S and ∃ t ∈ A s.t.(s,t) ∈ C} (35)

The value of this two frame interpretation of Dempster-Shafer Theory is

recognizing the contribution of DST as a new technique for propagating probabilities
through logical links, i.e. one can obtain information about one frame of discernment
from its logical relation to another frame. Specifically, if the only information available
between the elements of S and T (denoted by s and t, respectively) is through the logical
constraint (i.e., their compatibility relation C), traditional Bayesian theory has difficulty
providing for a meaningful inference regarding s and t. Dempster-Shafer theory can

23
represent the relationship, C, between s and t by a subset of the joint frame S×T. [Zhang,
1994]
With respect to the rule of combination, Zhang points out that Dempster’s rule
fails to consider how focal elements intersect. [Zhang, 1994] To define an alternative
rule of combination, he introduces a measure of the intersection of two sets A and B
assuming finite sets. This is defined as the ratio of the cardinality of the intersection of
two sets divided by the product of the cardinality of the individual sets. Zhang denotes
this relation with r(A,B):
A ∩ B C (36)
r( A, B ) = =
A B A B

where A ∩ B = C. The resulting combination rule scales the products of the basic
probability assignments of the intersecting sets (A ∩ B = C) by using a measure of
intersection, r(A,B) defined in Equation 36. This is repeated for every intersecting pair
that yields C. The scaled products of the masses for all pairs whose intersection equals C
are summed and multiplied by a factor k. In this case, k is a renormalization factor that is
independent of C, m1, and m2. This renormalization factor provides that the sum of the
basic assignments to add to 1.
m (C ) = k ∑ [
C
m 1 ( A) m 2 ( B)] (37)
A ∩B=C A B

The case where |C| = |A||B|, this rule will correspond to the Dempster rule.
It is important to note that the measure of intersection of two sets (r(A,B)) can be
defined in other ways, for example by dividing the cardinality of intersection of A and B
by the cardinality of the union of sets A and B. This would have the effect of a different
scaling on the product of the m’s that could be compensated for in the sum of all the basic
probability assignments by the renormalization factor k. Many combination rules could
be devised in the spirit of Zhang’s center combination rule by defining a reasonable
measure of intersection. This particular rule is commutative but not idempotent,
continuous, or associative.
The important contributions of Zhang’s work:

1. The two frame interpretation of Dempster-Shafer theory

2. The introduction of a measure of intersection of two sets (r(A,B)) based on
cardinality.
3. The center combination rule based on a measure of intersection of two sets
that could be modified by any other reasonable measure of intersection.

2.2.6: DUBOIS AND PRADE’S DISJUNCTIVE CONSENSUS RULE

Dubois and Prade take a set-theoretic view of a body of evidence to form their
disjunctive consensus rule in [Dubois, Prade, 1986; Dubois, Prade, 1992]. They define
the union of the basic probability assignments m1 ∪ m2 (denoted by m∪(C)) by extending
the set-theoretic union:
m ∪ ( C ) = ∑ A∪ B = C m 1( A )m 2 ( B ) (38)

24
For all A of the power set X. The union does not generate any conflict and does not reject
any of the information asserted by the sources. As such, no normalization procedure is
required. The drawback of this method is that it may yield a more imprecise result than
desirable.
The union can be more easily performed via the belief measure: Let Bel1∪Bel2 be
the belief measure associated with m1 ∪ m2. Then for every subset A of the universal set
X,
Bel1 ( A) ∪ Bel 2 ( A ) = Bel 1(A) Bel 2 (A) (39)

The disjunctive pooling operation is commutative, associative, but not idempotent.

2.2.7: MIXING OR AVERAGING

Mixing (or p-averaging or averaging) is a generalization of averaging for

probability distributions. [Ferson and Kreinovich, 2002] This describes the frequency of
different values within an interval of possible values in the continuous case or in the
discrete case, the possible simple events. The formula for the "mixing" combination rule
is just

1 n
m 1....n ( A) = ∑ wi mi ( A)
n i =1
(40)

where mi's are the bpa's for the belief structures being aggregated and the wi's are weights
assigned according to the reliability of the sources. This is very similar to the discount
and combine rule proposed by Shafer in that they are both averaging operations, but they
differ in which structures are being pooled. In the case of mixing, it is the basic
probability assignment, m; in the case of discount and combine, it is Bel.
Mixing generalizes the averaging operation that is usually used for probability
distributions. In particular, suppose that the input Dempster-Shafer structures are
probability distributions, that is, suppose that both structures consist of an element in
which each basic probability mass is associated with a single point. If one applies the
mixing operation to these inputs, the result will be a Dempster-Shafer structure all of
whose masses are also at single points. These masses and points are such that the
Dempster-Shafer structure is equivalent to the probability distribution that would have
been obtained by mixing the probability distributions, that is, by simply averaging the
probabilities for every point. None of the other Dempster-Shafer aggregation rules would
give this same answer. Insofar as averaging of probability distributions via mixing is
regarded as a natural method of aggregating probability distributions, it might also be
considered as a reasonable approach to employ with Dempster-Shafer structures, and that
is why it is considered here. Like mixing of probability distributions, mixing in
Dempster-Shafer theory is idempotent and commutative. It's not associative but it is
quasi-associative.

25
2.2.8: CONVOLUTIVE X-AVERAGING

Convolutive x-averaging (or c-averaging) is a generalization of the average for

scalar numbers. [Ferson and Kreinovich, 2002] This is given by the formula:

m12 (A) = ∑ m (B)m (C)

B+C
1 2 (41)
=A
2
Like the mixing average, this can be formulated to include any number of bpa’s, n, in the
following equation:
n
m 1.... n ( A) = ∑ ∏ m i ( Ai ) (42)
A1+ ... An
= A i =1
n

Suppose that the input Dempster-Shafer structures are scalar numbers, that is,
suppose that both structures consist of a single element where all mass is at a single point.
If one applies the convolutive average operation to these inputs, the result will be a
Dempster-Shafer structure all of whose mass is at a single point, the same point one gets
by simply averaging the two scalar numbers. None of the other Dempster-Shafer
aggregation rules would give this answer. Insofar as "averaging" is regarded as a natural
method of aggregating disparate pieces of information, it might also be considered as a
reasonable approach to employ with Dempster-Shafer structures, and that is why it is
considered here.
Like averaging of scalar numbers, the convolutive average is commutative. Also
like scalar averaging, the convolutive average is not associative, although it is quasi-
associative. Unlike scalar averaging, however, it is not idempotent.

2.2.9: OTHER RULES OF COMBINATION

There are still other rules of combination available for Dempster-Shafer theory
that will not be considered here. The remaining rules and the motivation for their
exclusion are summarized as follows:

Smets’ rule: Some authors refer to this as a distinct rule, however, this is essentially the
Dempster rule applied in Smets’ Transferable Belief Model. Smet’s model entails a
slightly different conception and formulation of Dempster-Shafer theory, though it
essentially distills down to the same ideas. [Smets, 2000]

Qualitative Combination Rule: This rule was proposed by Yao and Wong in their paper
[Yao and Wong, 1994]. This rule requires the definition of a binary relation expressing
the preference of one proposition or source, over another. Then a distance function is
defined between two belief relations. All the distances over all the pairs of the relation
are summed to obtain an overall distance. The resulting combination rule combines the
relations in such a way as to minimize the overall distance. This type of formulation of

26
DST, as its name implies is qualitative, whereas in engineering analyses, we expect to be
dealing with quantitative data. Consequently, it is beyond the scope of this study.

Yen’s rule: This rule is based on an extension of Dempster-Shafer theory by

randomizing the compatibility relations and using Zadeh’s relational model of Dempster-
Shafer theory. As this extension of DST is not the focus of the current paper and the rule
is similar to Zhang’s rule, a discussion of Yen’s rule is not included. [Yen, 1989]

Envelope, Imposition, and Horizontal x-Averaging: These are three methods of

combination that can be applied to belief structures that have been converted to
“generalized cumulative distribution functions” or p-boxes. The resultant combination
can be reinterpreted as a belief structure but with a complicated relationship with the
original inputs. A discussion of these methods in the context of p-boxes can be found in
[Ferson and Kreinovich, 2002].

3: DEMONSTRATION OF COMBINATION RULES

In this section, we demonstrate the differences between the various combination rules for
discrete and interval-type data. In Section 3.1, the data will be given by discrete values
and in Section 3.2 the data will be given by intervals.

3.1: Data given by discrete values

Suppose two experts are consulted regarding a system failure. The failure could
be caused by Component A, Component B or Component C. The first expert believes
that the failure is due to Component A with a probability of 0.99 or Component B with a
probability of 0.01 (denoted by m1(A) and m1(B), respectively). The second expert
believes that the failure is due to Component C with a probability of 0.99 or Component
B with a probability of 0.01 (denoted by m2(C) and m2(B), respectively). The
distributions can be represented by the following:

Expert 1:
m1(A) = 0.99 (failure due to Component A)
m1(B) = 0.01 (failure due to Component B)

Expert 2:
m2(B) = 0.01 (failure due to Component B)
m2(C) = 0.99 (failure due to Component C)

3.1.1: Dempster’s Rule

The combination of the masses associated with the experts is summarized in Table 1.

27
 Expert 1
A B C Failure
Cause
0.99 0.01 0 m1
Failure m2
Cause
A 0 m1(A) m2(A) = 0 m1(B) m2(A) m1(C) m2(A)
Expert =0 =0
2 B 0.01 m1(A) m2(B) = m1(B) m2(B) m1(C) m2(B)
0.0099 = 0.0001 =0
C 0.99 m1(A) m2(C) = m1(B) m2(C) m1(C) m2(C)
0.9801 = 0.0099 =0
Table 1: Dempster Combination of Expert 1 and Expert 2

Using Equations 11-13:

1. To calculate the combined basic probability assignment for a particular cell,

simply multiply the masses from the associated column and row.
2. Where the intersection is nonempty, the masses for a particular set from each
source are multiplied, e.g., m12(B) = (0.01)(0.01) = 0.0001.
3. Where the intersection is empty, this represents conflicting evidence and
should be calculated as well. For the empty intersection of the two sets A and
C associate with Expert 1 and 2, respectively, there is a mass associated with
it. m1(A) m2(C)=(0.99)(0.99) =(0.9801).
4. Then sum the masses for all sets and the conflict.
5. The only nonzero value is for the combination of B, m12(B) = 0.0001. In this
example there is only one intersection that yields B, but in a more complicated
example it is possible to find more intersections to yield B.
6. For K, there are three cells that contribute to conflict represented by empty
intersections. Using Equation 13, K = (0.99)(0.01) + (0.99)(0.01) +
(0.99)(0.99) = 0.9999
7. Using Equation 11, calculate the joint, m1(B) m2(B) = (.01)(.01) / [1-0.9999] =
1

Though there is highly conflicting evidence, the basic probability assignment for the
failure of Component B is 1, which corresponds to a Bel (B) = 1. This is the result of
normalizing the masses to exclude those associated with conflict. This points to the
inconsistency when Dempster’s rule is used in the circumstances of significant relevant
conflict that was pointed out by Zadeh.

28
3.1.2: Yager’s Rule

For this simple problem, Yager’s rule will yield the almost the same matrix as
with Dempster’s rule. However, there are some important exceptions in the
nomenclature and eventually the allocation of conflict:

1. Instead of basic probability assignments (m), Yager calls these ground

probability assignments (q)
2. Instead of using K to represent the conflict, Yager uses the q(Ø) which is
calculated in the exact same way as K. (Equation 13)

Using Equation 16, the combination is calculated:

q12(B) = m12(B) = (.01)(.01) = .0001

Here the combination is not normalized by the factor (1-K). When Yager converts the
ground probability assignments (q) to the basic probability assignments (m), the mass for
a particular joint remains the same and the mass associated with conflict is attributed to
the universal set X that represents the degree of ignorance (or lack of agreement). So in
this case the m(X) is 0.9999. To convert the basic probability assignment to the lower
bound Bel, the Bel(B) is equal to the m(B) (Bel (B) = .0001), as this is the only set that
satisfies the criteria for Belief (B ⊆B). This approach results in a significant reduction of
the value for Belief and a large expansion of Plausibility. Note that the value of Belief is
substantially smaller than either the experts’ estimates would yield individually and in
such a case, this may be counterintuitive.

3.1.3: Inagaki’s Rule

Once again the matrix is calculated in the same manner as in case of the Dempster
rule. Inagaki uses the ground probability functions similar to Yager. Ultimately, the
value of m12(B) obtained by Inagaki’s rule depends on the value of k which is now a
parameter. It is suggested by the literature that the value of k should be determined
experimentally or by expert expectation though an exact procedure is lacking. Figure 6
demonstrates the behavior of the Inagaki combination as a function of the value of k for
this problem.

29
 1.2

0.8
Inagaki m(B)

0.6

0.4

0.2

0
0
7
30
65
100
450
800

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
50
00
50
00
115
150
185
220
255
290
325
360
395
430
465
500
535
570
605
640
675
710
745
780
815
850
885
920
955
990
102
106
109
113
k

Figure 6: The value of m12(B) as a function of k in Inagaki’s rule

When k = 0, Inagaki’s combination will obtain the same result as Yager’s (m12(B)
1 1
= .0001). When k = = = 10000 , Inagaki’s rule corresponds to
1− q(∅) 1 − 0.9999
Dempster’s rule (m12(B) = 1). Because there is no mass associated with the universal set
q(X), in this case, Inagaki’s extra rule is the same as Dempster’s rule. Although, the
calculation can be extended beyond Dempster’s rule, any value for the combination
greater than 1 does not make sense because sums of all masses must be equal to 1.
Corresponding to the increasing value of k, is the increase in the filtering of the evidence.

3.1.4: Zhang’s Rule

Recall from Equations 36 and 37 for Zhang’s rule, in addition to calculating the
product of the masses like in Table 1, we must also calculate the measure of intersection
based on the cardinality of the sets. The cardinality of each of the sets A, B, and C is 1.
In this case we find that the only nonzero intersection of the sets is set B obtained from
the evaluation of B by both Experts 1 and 2. Since |B|=|B||B|, we find that the Zhang
combination corresponds to the Dempster combination. This points to two problems with
Zhang’s measure of intersection:

1. The equivalence with Dempster’s rule when the cardinality is 1 for all relevant
sets or when the |C|=|A||B| in the circumstance of conflicting evidence. (This
should not pose a problem if there is no significant conflict.)
2. If the cardinality of B was greater than 1, even completely overlapping sets
will be scaled.

3.1.5: Mixing

The formulation for mixing in this case corresponds to the sum of m1(B)(1/2) and
m2(B)(1/2). From Equation 40:

30
m12(A) = (1/2)(0.99) = 0.445

m12(B) = (1/2)(0.01)+ (1/2) (0.01) = 0.01

m12(C) = (1/2)(0.99) = 0.445

3.1.6: Dubois and Prade’s Disjunctive Consensus Pooling

The unions of multiple sets based on the calculations from Table 1 that can be
summarized in Table 2.

Union m∪ Linguistic Interpretation

A∪A 0 Failure of Component A
A∪B 0.0099 Failure of Component A or B
A∪C 0.9801 Failure of Component A or C
B∪B 0.0001 Failure of Component B
B∪C 0.0099 Failure of Component B or C
C∪C 0 Failure of Component C
A∪B ∪C 1 Failure of Component A or B or C
Table 2: Unions obtained by Disjunctive Consensus Pooling

3.2: Data given by intervals

Using the operations discussed above, now we will consider the aggregation of
three sources of information where the information is given as intervals. Interval-based
data is common to problems involving parametric uncertainty for physical parameters
like conductivity, diffusivity, or viscosity. Suppose there is an experiment that provides
multiple intervals for an uncertain parameter from three sources A, B, and C that must be
combined. The intervals associated with sources A, B, and C are summarized in the
Tables 3,4, and 5, respectively. Figures 7, 8, and 9 depict the intervals and the basic
probability assignments graphically with a “generalized cumulative distribution function”
(gcdf). This is the probabilistic concept of cumulative distribution function generalized
to Dempster-Shafer structures where the focal elements (intervals) are represented on the
x-axis and the cumulative basic probability assignments on the y-axis. A discussion of
the generalization of some of the ideas from the theory of random variable to the
Dempster-Shafer environment is discussed in [Yager, 1986].

Interval m1
[1,4] 0.5
[3,5] 0.5
Table 3: The interval-based data for A and the basic probability assignments

31
 1

0 .5

0
0 1 2 3 4 5 6

Figure 7: The gcdf of A

Interval m2
[1,4] 0.3333
[2,5] 0.3333
[3,6] 0.3333
Table 4: The interval-based data for B and the basic probability assignments

1
b

0 .5

0
0 1 2 3 4 5 6 7

Figure 8: The gcdf of B

32
 Interval m3
[6,10] 0.3333
[9,11] 0.3333
[12,14] 0.3333
Table 5: The interval-based data for C and the basic probability assignments

0 .5

0
5 1 5

Figure 9: The gcdf of C

Without any combination operation, the gcdf’s of A, B, and C are represented in Figure
10.
1

A
0.5
B

0
0 10 20
Figure 10: The gcdf’s of A, B, and C without any combination operation

As is evident in Figure 10 and Tables 3,4 and 5, the data for A and B is consistent
with each other. However the data for A and C are disjoint. First, we will consider the
combination of consistent data (A and B) and then the combination of the disjoint data (A
and C) with the combination rules discussed in Section 2.

33
3.2.1: Dempster’s Rule

The calculation of Dempster’s rule (Equation 11-13) is summarized in Table 6.

A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m
[1, 4] 0.33333 [1, 4] 0.16667 [3, 4] 0.16667

B [2, 5] 0.333333 [2, 4] 0.16667 [3, 5] 0.16667

[3, 6] 0.333333 [3, 4] 0.16667 [3, 5] 0.16667

Table 6: Combination of A and B with Dempster’s Rule

Note that the intersection of two intervals is defined by the maximum of the two lower
bounds and the minimum of the two upper bounds corresponding to an intersection. The
bpa's for like intervals are summed, i.e. [1,4] has a value for m of 0.166667; [2,4] has an
m value of 0.166667; [3,4] has a value of 0.33334; and [3,5] has an m value of 0.33334.
The resulting structure of the combination of A and B using Dempster’s rule is
depicted in Figure 11.

0.5

0
0 1 2 3 4 5 6

Figure 11: The gcdf of the combination of A and B using Dempster’s rule

The combination of A and C using Dempster’s rule is not possible due to the
normalization factor.

34
3.2.2: Yager’s Rule

As the evidence from A and B is consistent, the calculations for Yager’s rule are same as
in Table 6. The resulting structure of the combination of A and B using Yager’s rule
(Figure 12) is also the same as with Dempster’s rule.
1

0.5

0
0 1 2 3 4 5 6

Figure 12: The combination of A and B using Yager’s rule

Unlike the Dempster’s case, Yager’s rule can be calculated for the combination of A and
C. However, since the evidence is entirely conflicting, all of the basic probability mass is
attributed to the universal set. In the continuous domain this corresponds to the real line.
As noted earlier, the mass allocated to the universal set is interpreted as the degree of
ignorance or the degree of lack of agreement among sources.

0.5

0
-10 0 10 20 30

Figure 13: The gcdf of the combination of A and C using Yager’s rule

35
3.2.3: Inagaki’s Rule

Using k=0, we obtain the same calculations as Yager’s rule and Dempster’s rule.

0.5

0
0 1 2 3 4 5 6

Figure 14: The Inagaki combination of A and B for k=0

As expected, we find the same calculations for the combination of A and B where k=1.

0.5

0
0 1 2 3 4 5 6

Figure 15: The Inagaki combination of A and B where k = 1

3.2.4: Zhang’s Rule

For the step-by step calculation of Zhang’s combination rule, first take the masses
obtained by the simple product of the marginals.

36
 A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m 1 4 3 5
[1, 4] 0.333333 [1, 4] 0.166667 [3, 4] 0.166667
1 4 1 4 3 4
B [2, 5] 0.333333 [2, 4] 0.166667 [3, 5] 0.166667
2 5 2 4 3 5
[3, 6] 0.333333 [3, 4] 0.166667 [3, 5] 0.166667
3 6 3 4 3 5
Table 7: The combination of the marginals with Zhang’s rule

Next we calculate a measure of intersection. In the continuous case, we have

elected to interpret interval length for the calculation of the measure of intersection.

A
Interval A length Interval A length
[1, 4] 3 [3, 5] 2
Interval B length Interval A∩B Interval A∩B
length length
[1, 4] 3 [1, 4] 3 [3, 4] 1
B [2, 5] 3 [2, 4] 2 [3, 5] 2
[3, 6] 3 [3, 4] 1 [3, 5] 2
Table 8: The length of the intervals and their intersections

Then calculate the value of r(A,B) from Equation 36:

A
Interval A length Interval A length
[1, 4] 3 [3, 5] 2
Interval B length Interval r(A,B) Interval r(A,B)
[1, 4] 3 [1, 4] 0.333333 [3, 4] 0.166667
B [2, 5] 3 [2, 4] 0.222222 [3, 5] 0.333333
[3, 6] 3 [3, 4] 0.111111 [3, 5] 0.333333
Table 9: Calculation of the Measure of Intersection

Multiply the basic probability masses (m) from Table 8 by the r(a,B) in Table 9.

37
 A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m Interval r(A,B)*m Interval r(A,B)*m
[1, 4] 0.333333 [1, 4] 0.055556 [3, 4] 0.027778
B [2, 5] 0.333333 [2, 4] 0.037037 [3, 5] 0.055556
[3, 6] 0.333333 [3, 4] 0.018519 [3, 5] 0.055556
Table 10: The product of m and r(A,B)
The sum of all of the masses m, scaled by r(A,B) is 0.25. So the renormalization
factor k is the inverse of this sum, 4. All of the masses are then renormalized by
multiplying each by 4.

A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m Interval r(A,B)*m Interval r(A,B)*m
[1, 4] 0.333333 [1, 4] 0.222222 [3, 4] 0.111111
B [2, 5] 0.333333 [2, 4] 0.148148 [3, 5] 0.222222
[3, 6] 0.333333 [3, 4] 0.074074 [3, 5] 0.222222
Table 11: The renormalized masses with Zhang’s rule
Once again, masses for like intervals are summed to obtain the final distribution, i.e. [1,4]
has an m value of 0.22222; [2,4] has an m value of 0.14815; [3,4] has an m value of
0.18519; [3.5] has an m value of 0.44444.
These gcdf of the renormalized masses are graphed in Figure 16.

0.5

0
0 1 2 3 4 5 6

Figure 16: The Zhang combination of A and B

38
As can be seen in Figure 16 and its corresponding table (Table11) when compared to
those of the other Dempster-type rules (Dempster’s rule, Yager’s rule, and Inagaki’s rule
for k=0 and k=1), the Zhang rule yields a slightly different answer for the combination.
As there is no overlap between the two inputs, the combination of A and C is not
possible using Zhang’s rule.

3.2.5: Mixing

Using Equation 40, the values for mixing (without weights) are listed in Table 12:

Sources Initial Interval m Final Interval m

Source 1 [1, 4] 0.5 [1, 4] 0.25
[3, 5] 0.5 [3, 5] 0.25
Source 2 [1, 4] 0.333333 [1, 4] 0.166667
[2, 5] 0.333333 [2, 5] 0.166667
[3, 6] 0.333333 [3, 6] 0.166667

Table 12: The mixture of A and B

The masses for the like final intervals are summed: [1,4] has an m value of 0.41667; the
remaining distributions remain the same. The resulting structure of the combination of A
and B using mixing can be observed in the Figure 17.
1

0.5

0
0 1 2 3 4 5 6 7

Figure 17: The mixture of A and B

The combination of A and C is possible using mixing. These calculations are

summarized in Table 13.

39
 Sources Initial Interval m Final Interval m
Source 1 [1, 4] 0.5 [1, 4] 0.25
[3, 5] 0.5 [3, 5] 0.25
Source 2 [6, 10] 0.333333 [6, 10] 0.166667
[9, 11] 0.333333 [9, 11] 0.166667
[12, 14] 0.333333 [12, 14] 0.166667

Table 13: The mixture of A and C

The resulting structure of the combination of A and C using mixing:

0.5

0
0 10 20

Figure 18: The mixture of A and C

3.2.6: Convolutive x-Averaging

Using Equation 42, we calculate the convolutive x-average for A and B found in
Table 14.

A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m 1 4 3 5
[1, 4] 0.33333333 [1, 4] 0.16666667 [2, 4.5] 0.16666667
1 4 1 4 1 0.5
B [2, 5] 0.33333333 [1.5, 4.5] 0.16666667 [2.5, 5] 0.16666667
2 5 1.5 4.5 2 0.5
[3, 6] 0.33333333 [2, 5] 0.16666667 [3, 5.5] 0.16666667
3 6 3 0.5 3 0.5
Table 14: The Combination of A and B using Convolutive x-Averaging

40
 The resulting structure of the combination of A and B using convolutive x-
averaging is depicted in Figure 19.

0.5

0
0 1 2 3 4 5 6

Figure 19: The gcdf of the combination of A and B using convolutive x-averaging

To see the difference between the Dempster rule (solid line) and convolutive x-
averaging (dashed line) for the combination of A and B refer to Figure 20.
1

Convolutive
X-Avg.
0.5 Dempster

0
0 1 2 3 4 5 6

Figure 20: The Comparison of Combinations of A and B with Dempster’s rule and
Convolutive X-Averaging

As is readily apparent, the bound for the convolutive x-average either is equal to or is
significantly larger than the bounds of the Dempster combination.
The combination for A and C can be performed though the convolutive x-average
and the calculation are shown in Table 15.

41
 A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m 1 4 3 5
[6, 10] 0.33333333 [3.5, 7] 0.16666667 [4.5, 7.5] 0.16666667
6 10 3.5 7 6 0.5
C [9, 11] 0.33333333 [5, 7.5] 0.16666667 [6, 8] 0.16666667
9 11 5 7.5 9 0.5
[12, 14] 0.33333333 [6.5, 9] 0.16666667 [7.5, 9.5] 0.16666667
12 14 12 0.5 12 0.5
Table 15: The Combination of A and C using Convolutive x-Averaging

0.5

0
3 4 5 6 7 8 9 10

Figure 21: The gcdf of the Combination of A and C using Convolutive x-Averaging
The difference between the Yager rule and convolutive x-averaging for the combination
of A and C.

42
 1

0.5 Yager

Convolutive
X-Avg.

0
3 4 5 6 7 8 9 10

Figure 22: Comparison of Yager’s rule and Convolutive x-averaging for A and C
This dramatically demonstrates the difference between the Yager combination under
complete conflict (which corresponds to the whole real line) and the convolutive x-
average. Yager’s distribution implies that there is complete ignorance regarding the
inputs, whereas the convolutive x-average simply averages them and provides a
significantly narrower answer.

3.2.7 Dubois and Prade’s Disjunctive Consensus

The upper and lower bounds for the disjunctive consensus are defined by the
minimum of the lower bounds and the maximum of the upper bounds. The calculations
for the joint of the basic probability assignments is the product of the marginals. This is
also known as a convex hull of all unions. The intervals and their respective probability
assignments are listed in Table 16.

A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m 1 4 3 5
[1, 4] 0.333333 [1, 4] 0.166667 [1, 5] 0.166667
1 4 1 4 1 5
B [2, 5] 0.333333 [1, 5] 0.166667 [2, 5] 0.166667
2 5 1 5 2 5
[3, 6] 0.333333 [1, 6] 0.166667 [3, 6] 0.166667
3 6 1 6 3 6
Table 16: The Disjunctive Consensus Pooling of A and B

The only like interval is [1,5] where the summed m is equal to 0.33334. The other
distribution remain the same as in Table 16.

43
 1

0.5

0
0 1 2 3 4 5 6 7

Figure 23: The Disjunctive Consensus Pooling of A and B

It is also possible to calculate the combination of A and C using disjunctive consensus
pooling.

A
Interval m Interval m
[1, 4] 0.5 [3, 5] 0.5
Interval m 1 4 3 5
[6, 10] 0.333333 [1, 10] 0.166667 [3, 10] 0.166667
6 10 1 10 3 10
C [9, 11] 0.333333 [1, 11] 0.166667 [3, 11] 0.166667
9 11 1 11 3 11
[12, 14] 0.333333 [1, 14] 0.166667 [3, 14] 0.166667
12 14 1 14 3 14
Table 17: Calculations for the Disjunctive Consensus Pooling of A and C
1

0.5

0
0 10 20

Figure 24: The gcdf for the Disjunctive Consensus Pooling of A and C

See [Ferson and Kreinovich, 2002] for a comparison of the disjunctive consensus and the
envelope operation.

44
3.2.8: Summary of Examples

A simple comparison of the combinations of A and B and A and C with the

various rules is summarized in Table 18 and Table 19:

COMBINATION COMMENTS
RULES
Dempster’s Rule The intervals are defined by the minimum of the upper bounds
and maximum of the lower bounds. The individual bpa’s are
calculated by multiplying the bpa’s of the marginals. Where the
same interval is obtained from multiple combinations, the
associated bpa’s are summed. No normalization step is taken in
this example, as there is no mass associated with conflict.
Yager’s Rule As there is no conflict, this problem provides the same answer as
Dempster’s rule.
Inagaki’s Rule (k=0) As there is no conflict, this problem provides the same answer as
Dempster’s rule.
Inagaki’s Rule (k=1) As there is no conflict, this problem provides the same answer as
Dempster’s rule.
Zhang’s Rule Provides a slightly different answer than the other Dempster-type
rules. The intervals are defined in the same manner but the bpas
are scaled differently because of the measure of intersection.
Consequently some bpa’s are larger than those obtained by
Dempster’s rule, while others are slightly smaller. The final
masses are renormalized so all masses will add to one.
Mixing This averaging operation provides different intervals and different
bpa’s than Dempster’s rule. The intervals are either equal to the
Dempster intervals or in most cases wider. The bpa’s are more
concentrated on the interval [1,4].
Convolutive The convolutive x-average is quite different from Dempster’s
x-Average rule, Zhang’s rule, and mixing in terms of the bounds of the
interval and their respective bpa’s. The bounds of this average
are either equal to those of Dempster’s rule or larger.
Disjunctive As expected, this is by far the most imprecise of the combination
Consensus Pooling methods. The intervals are defined by the maximum of the upper
bounds and the minimum of the lower bounds and the bpa’s are
calculated in the same manner as Dempster’s rule. Consequently,
in this example, this method provides fewer intervals than in
Dempster’s rule which are either equal to or greater than the
Dempster intervals.
Table 18: The Combination of A and B Comparison Table

45
 COMBINATION COMMENTS
RULES
Dempster’s Rule No answer is possible.
Yager’s Rule To reflect the complete conflict between the two sources, Yager’s
rule provides the universal set or the real line as its answer.
Inagaki’s Rule (k=0) Provides the same answer as Yager’s rule, i.e., the universal set or
the real line.
Inagaki’s Rule (k=1) No answer is possible.
Zhang’s Rule No answer is possible.
Mixing The mixture maintains the same intervals as the inputs but divides
the bpa by 2, the number of sources. While this does provide an
answer, the issue of conflict is not represented. The gcdf reflects
the full scope of the input bounds.
Convolutive The convolutive x-average provides different intervals than
x-Average obtained by mixing. The upper bounds of the marginals are
averaged to obtain the upper bound of the joint. The same
process is repeated for the lower bound. The bpa’s are the
product of the marginal’s masses. Consequently, this average is
different than the mixing average and the gcdf is concentrated in
the center of the two inputs.
Disjunctive As a union operation, this finds the largest possible intervals
Consensus Pooling obtained by the two inputs and calculates the joint bpa’s by
multiplying the marginal bpa’s. The answer subsumes both
answers provided by mixing and the convolutive x-average.

Table 19: The Combination of A and C Comparison Table

As indicated in Table 19, when the sources are completely conflicting, some rules
will not apply at all (Dempster rule, Zhang’s rule) or provide an answer that corresponds
to complete ignorance (Yager’s rule). The averaging operations will work but it may be
inappropriate to average two extremes to produce an answer that neither source suggested
was a possible answer.

4: CONCLUSIONS

Dempster-Shafer Theory essentially combines the Bayesian notion of

probabilities with the classical idea of sets where a numerical value signifying confidence
can be assigned to sets of simple events rather than to just mutually exclusive simple
events. [Bogler, 1992] The theoretical basis for Dempster-Shafer Theory is an attractive
one for dealing with a corpus of data that requires different degrees of resolution. From
the operational perspective of Dempster-Shafer theory, we find that the aggregation of
evidence from multiple sources is not straightforward, as there are a variety of possible
combination rules.
As there are multiple ways of combining data, it would be desirable to develop a
formal procedure by which one could select an appropriate combination operation.
Although the algebraic properties may not prove to be useful in designing a

46
comprehensive typology of combination operators, they do provide insight into some of
the behavior of the operators. Some of the algebraic properties of the combination rules
discussed in this report are summarized in Table 20.

Algebraic Properties
Combination Idempotent Commutative Associative Quasi-
Rules Assocative
Dempster’s No Yes Yes
Rule
Yager’s Rule No Yes No Yes
Inagaki’s Rule No Yes Depends on Depends
value of k on value of
k
Zhang’s Rule No Yes Yes
Mixing Yes Yes Yes
Convolutive Yes Yes Yes
x-Average
Disjunctive No Yes Yes
Consensus
Pooling

Table 20: Combination Rules and Their Algebraic Properties

For Dubois and Prade, combination operations cannot be discussed solely in terms
of algebraic properties because the imposition of too many properties can be too
restrictive to solve practical problems. As we can see with the numerous Dempster-type
combination rules, they satisfy many of the same algebraic properties. Moreover, as the
work of Dubois and Prade points out, [Dubois, Prade, 1992], even the definitions of the
algebraic properties can be problematic and debatable. Nevertheless, understanding what
are the desirable properties of a prospective combination rule can be one part of the
criteria for rule selection.
Another helpful heuristic for choosing a combination rule is to identify the
requirements of the situation as disjunctive pooling, conjunctive pooling or tradeoff.
These correspond to the Dubois and Prade disjunctive pooling method, the Dempster
rule, and the remaining operations of Yager’s rule, Zhang’s rule, discount and combine,
mixing, and convolutive x-averaging, respectively. If that requirement alone cannot be
determined, it may prove practical to apply Inagaki’s rule for many values of k. As we
have shown here, a number of these rules can be tested and their results compared. Many
of the Dempster and the “Dempster-type” combination rules share a common first step,
the multiplication of the marginal masses to find the joint. These rules fundamentally
differ on how these joint masses are to be combined and where to allocate the mass
associated with conflict in the second step.
There are a number of considerations that need to be addressed when combining
evidence in Dempster-Shafer theory. Generally speaking, these include the evidence

47
itself, the sources of information, the context of the application, and the operation used to
combine the evidence. These are depicted in Figure 25.

COMBINATION
CONTEXT OPERATION
Relevance of Conflict Type of operation
Algebraic properties
Handling of conflict
EVIDENCE Advantages
Type of evidence
Disadvantages
Amount of evidence
Accuracy of evidence
SOURCES
Type of source
Number of sources
Reliability of sources
Dependency between
sources
Conflict between sources

Figure 25: Important Issues in the Combination of Evidence

As the literature survey on aggregation in generalized information theory reflects,

much of the research in the combination rules in Dempster-Shafer theory is devoted to
advancing a more accurate mathematical representation of conflict. In Figure 25, all of
the contextual considerations like the type, amount, and accuracy of evidence as well as
the type and reliability of sources and their interdependencies can be interpreted as
features of conflict assessment. Once values are established for degree of conflict, the
most important consideration is the relevance of the existing conflict. Though conflict
may be present, it may not always be contextually relevant. Take a target identification
problem where there are two sensors with a small overlapping area in their respective
ranges and the ultimate task is to assign priority to all detected targets. In this case, it is
intuitive to assign the highest priority to the target with the largest amount of mass
associated with it. We are not concerned with the mass allocated to other targets and
hence, conflict is not relevant in this case. Consequently, even in a context of highly
conflicting evidence, Dempster's rule might be the most appropriate rule to use as conflict
is normalized out of the combination if that conflict is determined by context to be
irrelevant. Dempster's rule allows for the comparative assessment the masses associated
with various targets independent of their location inside or outside of the intersection of
the two overlapping sets.
In conclusion of the discussion of the rules of combination in Dempster-Shafer
theory we find that under situations of minimal conflict or irrelevant conflict and all of
the sources can be considered reliable, a Dempster combination might be justified. As
was demonstrated by the example (Section 3.2.1), when there is a situation of no conflict,
two of the Dempster-type rules (Yager, Inagaki (k=0, k=1), provide the same answer as
Dempster’s rule. As the level of relevant conflict increases, Yager’s rule might more

48
appropriate as the conflict is not ignored. An advantage of Yager’s rule is that it
represents the level conflict by the basic probability assignment of the universal set X.
However, there is the possibility with Yager’s rule that the basic probability mass
associated with the combined result is significantly smaller than those provided by the
original sources (demonstrated in Section 3.1). Inagaki’s unified combination rule
investigates the effects of many different values for conflict on a combined result and
includes both Dempster’s rule and Yager’s rule. However, the procedure for the
contextual determination of the value of k for Inagaki’s rule is an important question that
is not clearly described in this current literature survey. Zhang’s rule provides a result for
the bpa of the combination that is scaled by a measure of the intersection but under
certain circumstances this measure can correspond to Dempster’s rule and suffer the same
criticisms under significant relevant conflict. If Yager’s rule begins to reflect a high level
of ignorance, the propriety of a combining the evidence at all should be considered. If a
combination is appropriate, possible methods for this case could be disjunctive consensus
pooling, the discount and combine method (when there is a qualified analyst to discount
based on source reliability), or other averaging methods like mixing or convolutive x-
averaging. With all the issues that have been discussed in this report with respect to the
combination of evidence in Dempster-Shafer theory, we find that most are linked to the
characterization of conflict. Consequently, we identify this as the most critical concern
for the specific selection of a combination operation. Specifically, what is the degree and
contextual relevance of conflict and how is this handled by a particular combination rule.

49
 REFERENCES

Ayyub, B. M. (2001). Elicitation of Expert Opinions for Uncertainty and Risks, CRC
Press.

Bogler, P. L. (1987). “Shafer-Dempster Reasoning with Applications to Multisensor

Target Identification Systems.” IEEE Transactions on Systems, Man, and
Cybernetics SMC-17(6): 968-977.

Chokr, B and V. Kreinovich. (1994). How far are we from complete knowledge?
Complexity of knowledge acquisition in the Dempster-Shafer approach.
Advances in the Dempster-Shafer Theory of Evidence. R. R. Yager, J. Kacprzyk
and M. Fedrizzi. New York, John Wiley & Sons, Inc.: 555-576

Dempster, A. P. (1967). “Upper and Lower Probabilities Induced by a Multivalued

Mapping.” The Annals of Statistics 28: 325-339.

Dubois, D. and H. Prade (1986). "A Set-Theoretic View on Belief Functions: Logical
Operations and Approximations by Fuzzy Sets." International Journal of General
Systems 12: 193-226.

Dubois, D. and H. Prade (1992). "On the combination of evidence in various

mathematical frameworks." Reliability Data Collection and Analysis. J. Flamm
and T. Luisi. Brussels, ECSC, EEC, EAFC: 213-241.

Ferson, S. and V. Kreinovich. (2002) "Representation, Propagation, and Aggregation of

Uncertainty." SAND Report [in progress]

Hacking, I. (1975) The Emergence of Probability. New York, Cambridge University

Press.

Hasegawa, S. and T. Inagaki (1994). Dempster-Shafer theoretic design of a decision

support system for a large-complex system. 3rd IEEE International Workshop on
Robot and Human Communication (RO-MAN), Nagoya, Japan.

Helton, J. C. (1997). “Uncertainty and Sensitivity Analysis in the Presence of Stochastic

and Subjective Uncertainty.” Journal of Statistical Computation and Simulation
57: 3-76.

Inagaki, T. (1991). “Interdependence between Safety-Control Policy and Multiple-Sensor

Schemes Via Dempster-Shafer Theory.” IEEE Transactions on Reliability 40(2):
182-188.

Klir, G. (1989). “Is There More to Uncertainty Than Some Probability Theorists Might
Have Us Believe?” International Journal General Systems 15: 347-378.

50
Klir, G. J. and R. M. Smith (2001). “On Measuring Uncertainty and Uncertainty-Based
Information: Recent Developments.” Annals of Mathematics and Artificial
Intelligence 32: 5-33.

Klir, G. J. and M. J. Wierman (1998). Uncertainty-Based Information: Elements of

Generalized Information Theory. Heidelberg, Physica-Verlag.

Kramosil, I (2001). Probabilistic Analysis of Belief Functions. New York, Kluwer

Academic/Plenum Publishers.

Oberkampf, W., J.Helton, and Kari Sentz (2001) "Mathematical Representations of

Uncertainty." AIAA Conference Proceedings: Non-Deterministic Approaches
Forum, (AIAA2001-1645), Seattle, WA April, 16-19, 2001.

Oblow, E.M. (1987). "O-Theory---A Hybrid Uncertainty Theory." International Journal

of General Systems 13: 95-106.

Savage, L.J. (1972) The Foundations of Statistics. New York, Dover Publications.

Shafer, G. (1976). A Mathematical Theory of Evidence. Princeton, NJ, Princeton

University Press.

Shafer, G. (1986). Probability Judgement in Artificial Intelligence. Uncertainty in

Artificial Intelligence. L. N. Kanal and J. F. Lemmer. New York, Elsevier
Science. 4.

Smets, P. (2000). Data Fusion in the Transferable Belief Model. Third International
Conference on Information Fusion, Paris, France.

Tanaka, K. and G. J. Klir (1999). “A design condition for incorporating human

judgement into monitoring systems.” Reliability Engineering and System Safety
65(3): 251-258.

Yager, R. (1986). “Arithmetic and other operations on Dempster Shafer structures.”

International Journal of Man-Machine Studies 25: 357-366.

Yager, R. (1987a). “On the Dempster-Shafer Framework and New Combination Rules.”
Information Sciences 41: 93-137.

Yager, R. R. (1987b). “Quasi-Associative Operations in the Combination of Evidence.”

Kybernetes 16: 37-41.

Yager, R. R., J. Kacprzyk, et al. (1994). Advances in the Dempster-Shafer Theory of

Evidence. New York, John Wiley & Sons, Inc.

51
Yao, Y. Y. and S. K. M. Wong (1994). “Representation, Propagation and Combination of
Uncertain Information.” International Journal of General Systems 23: 59-83.

Yen, J. (1989). "Can Evidence Be Combined in the Dempster-Shafer Theory."

Uncertainty in Artificial Intelligence 3. L. N. Kanal and J. F. Lemmer. New York,
Elsevier Science.

Zadeh, L. A. (1984). “Review of Books: A Mathematical Theory of Evidence.” The AI

Magazine 5(3): 81-83.

Zadeh, L. A. (1986). A Simple View of the Dempster-Shafer Theory of Evidence and its
Implication for the Rule of Combination. The AI Magazine. 7: 85-90.

Zhang, L. (1994). Representation, independence, and combination of evidence in the

Dempster-Shafer theory. Advances in the Dempster-Shafer Theory of Evidence.
R. R. Yager, J. Kacprzyk and M. Fedrizzi. New York, John Wiley & Sons, Inc.:
51-69.

52
 APPENDIX A
References to Applications of Dempster-Shafer Theory

SUBJECT Related Subject Headings Pages

Cartography Geography, Map building, Image Processing A-2
Classification, Pattern Recognition, Speaker Identification, A-2 to A-7
Identification, Recognition Voice Recognition, Decision-Making,
Radar, Target Identification, Optimization,
Fault Detection, Artificial Vision, Image
Processing, Multiple Sensors
Decision-Making Classification, Identification, Recognition, A-7 to
Risk Management, Expert Systems, Image A-11
Processing, Robotics
Engineering and Expert Systems, Decision Making A-11 to
Optimization A-12
Expert Systems Knowledge-based Systems, Identification, A-13 to
Fault Diagnosis, Geography, Control A-14
Systems, Decision-Making
Fault Detection and Failure Identification, Risk, Reliability, A-14 to
Diagnosis Classification, Sensors A-15
Image Processing Object Recognition, Expert Systems, A-16 to
Geography, Cartography, Radar, Target A-21
Identification, Biomedical Engineering
Medical Applications Expert Systems, Image Processing, Control A-21 to
Systems A-23
Miscellaneous Databases, Autonomous Vehicle A-23 to
Navigation, Expert Systems, Forecasting, A-27
Finance, Manufacturing, Document
Retrieval, Simulation, Decision-Making,
Climatology, Expert Opinion Pooling,
Optimization
Multiple Sensors Autonomous Vehicles, Target Identification, A-27 to
Pattern Recognition, Classification, A-29
Simulation, Artificial Vision, Satellites,
Robotics
Risk and Reliability Fault Diagnosis, Expert Systems, Decision- A-30
Making
Robotics Sensors, Decision-Making, Target A-31 to
Identification, Artificial Vision A-33
Signal Processing Sensors, Target Identification, Recognition, A-33 to
Classification, Radar, Detection, Expert A-34
Systems, Sensitivity Analysis

Appendix A-1
 CARTOGRAPHY
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Binaghi, E., L. “Slope instability zonation: Natural Hazards 17(1): 77- Cartography, This paper presents a comparison between two methodologies for the
Luzi, et al. A comparison between 97 (1998). Instability Maps, evaluation of slope instability and the production of instability maps,
certainty factor and fuzzy Evaluation using a probabilistic approach and a hybrid possibilistic and credibilistic
Dempster-Shafer approach. The first is the Certainty Factor method, and the second is
approaches.” based on Fuzzy Logic integrated with the Dempster-Shafer theory.

Leduc, F., B. “Combination of fuzzy sets Proceedings of SPIE The Cartography, Forest This paper explains a new approach to change detection and
Solaiman, et al. and Dempster-Shafer International Society for Map Updating interpretation in a context of forest map updating. The analysis of
theories in forest map Optical Engineering 4385: remotely sensed data always necessitates the use of approximate
updating using 323-335, (2001). reasoning. For this purpose, we use fuzzy logic to evaluate the objects'
multispectral data.” membership values to the considered classes and the Dempster-Shafer
theory to analyse the confusion between classes and to find the more
evident class to which an object belongs.

Tirumalai, A. P., “Evidential Reasoning for IEEE Transactions on Cartography, We address the problem of building a map of the environment utilizing
B. G. Schunck, et Building Environment Systems Man and Environmental sensory depth information obtained from multiple viewpoints. We
al. Maps.” Cybernetics 25(1): 10-20, Science present an approach for multi-sensory depth information assimilation
(1995). based on Dempster-Shafer theory for evidential reasoning.

CLASSIFICATION, IDENTIFICATION, AND RECOGNITION

AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Altincay, H. and “Novel rank-based ICASSP, IEEE Classification, In this paper, we propose a novel rank-based classifier combination
M. Demirekler classifier combination International Conference on Speaker scheme under uncertainty for speaker identification (SI). The
scheme for speaker Acoustics, Speech and Identification, combination is based on a heuristic method that uses Dempster-Shafer
identification.” Signal Processing theory of evidence under some conditions.
Proceedings 2: 1209-1212
(2000).

Bauer, M. “A Dempster-Shafer User Modeling and User- Plan Recognition, In this paper, an approach to the quantitative modeling of the required
Approach to Modeling Adapted Interaction 5(3-4): Modeling agent-related data and their use in plan recognition is presented. It relies
Agent Preferences for Plan 317-348 (1995). on the Dempster-Shafer Theory and provides mechanisms for the
Recognition.” initialization and update of corresponding numerical values.

Appendix A-2
 CLASSIFICATION, IDENTIFICATION, AND RECOGNITION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Chibelushi, C. C., “Audio-visual person IEE Conference Person Audio-visual person recognition promises higher recognition accuracy
F. Deravi, et al. recognition: An evaluation Publication(437): 26-30, Recognition, than recognition in either domain in isolation. To reach this goal, special
of data fusion strategies.” (1997). Multimedia, attention should be given to the strategies for combining the acoustic and
Decision-Making visual sensory modalities. This paper presents a comparative assessment
of three decision-level data fusion techniques for person identification:
Bayesian, Dempster-Shafer and possiblistic approaches.

Dekorvin, A., V. “Using Multiple Sources of Stochastic Analysis and Object recognition, The authors discuss an object recognition problem in which the
Espino, et al. Information to Recognize Applications 10(5): 573- Classification, characteristic features of the object are reported by remote sensors. We
and Classify Objects.” 589 (1992). Identification then extend the method to a more general class of selection problems and
consider several different scenarios. Fuzzy sets are used to represent
vague information. Information from independent sources is combined
using the Dempster-Shafer approach adapted to the situation in which the
focal elements are fuzzy as in the recent paper by J. Yen.

Dekorvin, A., R. “The Object Recognition International Journal of Object The goal of the present work is to obtain a reasonable solution to the
Kleyle, et al. Problem When Features Approximate Reasoning Recognition, problem of object identification. Sensors report on certain independent
Fail to Be Homogeneous.” 8(2): 141-162, (1993). Identification, feature values of an object. The Dempster-Shafer theory is used to
Classification integrate the information coming from these independent sources.

Denoeux, T. “Evidence-theoretic neural Pattern A new classifier based on the Dempster-Shafer theory of evidence is
network classifier.” Recognition, presented. The approach consists in considering the similarity to
Classification prototype vectors as evidence supporting certain hypotheses concerning
the class membership of a pattern to be classified. The different items of
evidence are represented by basic belief assignments over the set of
classes and combined by Dempster's rule of combination.

Denoeux, T. “K-nearest neighbor IEEE Transactions on Classification In this paper, the problem of classifying an unseen pattern on the basis of
classification rule based on Systems, Man and its nearest neighbors in a recorded data set is addressed from the point of
Dempster-Shafer theory.” Cybernetics 25: 804-813 view of Dempster-Shafer theory. Each neighbor of a sample to be
(1995). classified is considered as an item of evidence that supports certain
hypotheses regarding the class membership of that pattern.

Appendix A-3
 CLASSIFICATION, IDENTIFICATION, AND RECOGNITION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Denoeux, T. “Function approximation IEEE International Functional We propose a novel approach to functional regression based on the
in the framework of Conference on Neural Regression, Transferable Belief Model, a variant of the Dempster-Shafer theory of
evidence theory: A Networks Conference Prediction evidence. This method uses reference vectors for computing a belief
connectionist approach.” Proceedings 1: 199-203 structure that quantifies the uncertainty attached to the prediction of the
(1997). target data, given the input data.

Denoeux, T. “Reasoning with imprecise International Journal of Pattern This paper extends the theory of belief functions by introducing new
belief structures.” Approximate Reasoning Classification concepts and techniques, allowing to model the situation in which the
20(1): 79-111, (1999). beliefs held by a rational agent may only be expressed (or are only
known) with some imprecision. Central to our approach is the concept of
interval-valued belief structure (IBS), defined as a set of belief structures
verifying certain constraints. An application of this new framework to the
classification of patterns with partially known feature values is
demonstrated.

Denoeux, T. “A neural network IEEE Transactions on Pattern A new adaptive pattern classifier based on the Dempster-Shafer theory of
classifier based on Systems Man and Classification evidence is presented. This method uses reference patterns as items of
Dempster-Shafer theory.” Cybernetics Part a-Systems evidence regarding the class membership of each input pattern under
and Humans 30(2): 131- consideration.
150, (2000).
Denoeux, T. and “Handling possibilistic Fuzzy Sets and Systems Pattern A category of learning problems in which the class membership of
L. M. Zouhal labels in pattern 122(3): 409-424, (2001). Classification, training patterns is assessed by an expert and encoded in the form of a
classification using Decision-Making possibility distribution is considered. Two approaches are proposed,
evidential reasoning.” based either on the transformation of each possibility distribution into a
consonant belief function, or on the use of generalized belief structures
with fuzzy focal elements. In each case, a belief function modeling the
expert's beliefs concerning the class membership of each new pattern is
obtained.

Dillard, R. A. “Tactical Inferencing with Conference Record

the Dempster-Shafer Asilomar Conference on
Theory of Evidence.” Circuits, Systems &
Computers 17th: 312-316,
(1984).

Appendix A-4
 CLASSIFICATION, IDENTIFICATION, AND RECOGNITION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Foucher, S., J. M. “Multiscale and IEEE International Classification, We propose to use evidential reasoning in order to relax bayesian
Boucher, et al. multisource classification Conference on Image Radar decisions given by a multiscale markovian classification algorithm
using Dempster-shafer Processing 1: 124-128, (ICM). The Dempster-shafer rule of combination enables us to fuse
theory.” (1999). decisions in a local spatial neighbourhood which we further extend to be
multiscale and multisource. This approach enables us to more directly
fuse multiscale information. Application to the classification of very
noisy radar images produce interesting results.

Gang, T. and L. “Technique of multi-source Jixie Gongcheng Recognition, According to the Dempster-Shafer theory, the information fusion method
Wu information fusion and Xuebao/Chinese Journal of Decision-Making and classification decision strategies in ultrasonic detection were studied.
defects recognition in Mechanical Engineering 35: On this basis, the primary experimental research on the classification and
ultrasonic detection.” 11-14, (1999). recognition of the defects based on the information fusion has been
carried out.
Horiuchi, T. “Decision rule for pattern IEEE Transactions on Pattern In this paper, a pattern classification theory using feature values defined
classification by Pattern Analysis and Classification, on closed interval is formalized in the framework of Dempster-Shafer
integrating interval feature Machine Intelligence 20(4): Decision-Making measure. Then, in order to make up lacked information, an integration
values.” 440-448, (1998). algorithm is proposed, which integrates information observed by several
information sources with considering source values.

Kawade, M. “Object recognition system IEEE International Object Recognition In this paper, we propose an object recognition system in a dynamic
in a dynamic Conference on Fuzzy environment based on fuzzy logic and Dempster-Shafer's Theory which
environment.” Systems 3: 1285-1290, can integrate various inferences.
(1995).
Khalaf, S., P. Siy, “2-D and 3-D touching part Proceedings IEEE Recognition, A unified approach is presented for solving the 2-D and 3-D touching
et al. recognition using the International Symposium on Identification, part recognition problem. The problem is formulated as a Dempster-
theory of evidence.” Circuits and Systems 2: 992- Decision-Making Shafer evidence accumulation process.
994, (1990).

Appendix A-5
 CLASSIFICATION, IDENTIFICATION, AND RECOGNITION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Nigro, J. M., S. “Driving situation IEEE Conference on Recognition, We interested in the recognition of the maneuvers performed by the
Loriette recognition in the Intelligent Transportation Driving Maneuver, driver, specially the overtaking maneuver. We consider a maneuver as a
Rougegrez, et al. CASSICE project towards Systems, Proceedings, Sensors sequence of events. Then, according to the inputs obtained from the
an uncertainty ITSC: 71-76, (2000). system's sensors at different times, the goal is to evaluate the confidence
management.” of which particular maneuver is in progress. In this paper, the confidence
is modeled by a distribution of mass of evidence as proposed in the
Dempster-Shafer's theory.

Peddle, D. R. “Knowledge Formulation Photogrammetric Classification, Land The Dempster-Shafer Theory of Evidence provides an appropriate
for Supervised Evidential Engineering and Remote Cover framework for overcoming problems associated with the analysis,
Classification.” Sensing 61(4): 409-417, Classification integration, and classification of modern, multisource data sets. However,
(1995). current methods for generating the prerequisite evidence are subjective
and inconsistent. To address this, a more objective approach is presented
for deriving evidence from histogram bin transformations of supervised
training data frequency distributions. The procedure is illustrated by an
example application in which evidential land-cover classification.

Vasseur, P., C. “Perceptual organization Pattern Recognition 32(8): Pattern In this paper, we propose an application of the perceptual organization
Pegard, et al. approach based on 1449-1462, (1999). Recognition, based on the Dempster-Shafer theory. This method is divided into two
Dempster-Shafer theory.” Object parts which rectify the segmentation mistakes by restoring the coherence
Recognition, of the segments and detects objects in the scene by forming groups of
Identification, primitives. We show how we apply the Dempster-Shafer theory, usually
Optimization used in data fusion, in order to obtain an optimal adequation between the
perceptual organization problem and this tool.

Xu, L., A. “Methods of Combining IEEE Transactions on Classification, Method of combining the classification powers of several classifiers is
Krzyzak, et al. Multiple Classifiers and Systems, Man, and Pattern Recognition regarded as a general problem in various applications areas of pattern
Their Applications to Cybernetics 22(3): 418-435, recognition, and a systematic investigation has been made. Possible
Handwriting Recognition.” (1992). solutions to the problem can be divided into three categories according to
the levels of information available from the various classifiers. Four
approaches are proposed based on different methodologies for solving
this problem.

Appendix A-6
 CLASSIFICATION, IDENTIFICATION, AND RECOGNITION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Zhu, D. P., R. W. “A prototype vision system IEEE Transactions on Artificial Vision, To fully optimize the value of material produced from a hardwood log
Conners, et al. for analyzing CT imagery Systems Man and Image, Defect requires information about type and location of internal defects in the
of hardwood logs.” Cybernetics Part B- Detection, log, This paper describes a prototype vision system that automatically
Cybernetics 26(4): 522-532, Classification, locates and identifies certain classes of defects in hardwood logs. To
(1996). Object further help cope with the above mentioned variability, the Dempster-
Identification Shafer theory of evidential reasoning is used to classify defect objects.

DECISION-MAKING
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Attoh-Okine, N. “Use of belief function in Journal of Urban Planning Decision-Making, The Dempster-Shafer theory of combination is used to combine
O. and J. Gibbons brownfield infrastructure and Development-Asce Urban independent evidence from various issues to determine the overall
redevelopment decision 127(3): 126-143 (2001). Development uncertainty in redevelopment decision-making.
making.”

Bauer, M. “Approximation algorithms International Journal of Decision-Making This article reviews a number of algorithms based on a method of
and decision making in the Approximate Reasoning simplifying the computational complexity of DST.
Dempster-Shafer theory of 17(2-3): 217-237 (1997).
evidence : An empirical
study.”

Beynon, M., D. “An expert system for Expert Systems with Decision-Making This paper outlines a new software system we have developed that
Cosker, et al. multi-criteria decision Applications 20(4): 357- utilises the newly developed method (DS/AHP) which combines aspects
making using Dempster 367 (2001). of the Analytic Hierarchy Process (AHP) with Dempster-Shafer Theory
Shafer theory.” for the purpose of multi-criteria decision making (MCDM).

Beynon, M., B. “The Dempster-Shafer Omega-International Decision-Making We discuss recent developments of Dempster-Shafer theory including
Curry, et al. theory of evidence: an Journal of Management analytical and application areas of interest. We discuss developments via
alternative approach to Science 28(1): 37-50 the use of an example incorporating DST with the Analytic Hierarchy
multicriteria decision (2000). Process (AHP).
modelling.”

Appendix A-7
 DECISION-MAKING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Bharadwaj, K. “Hierarchical Censored Decision-Making The Dempster-Shafer Theory is used to formalize Variable Precision
K., Neerja, et al. Production Rules (Hcprs) Logic (VPL) type inference provides a simple, intuitive notion of the
System Employing the precision of an inference which relates it to the amount of information
Dempster-Shafer found. This formalism allows the ignorance in the evidence to be
Uncertainty Calculus.” preserved through the reasoning process and expressed in the decision.

Bosse, E. and J. “Fusion of identity Optical Engineering 36(3): Object The problem of fusing identity declarations emanating from different
Roy declarations from 648-657 (1997). Identification, sources is explored and decision makers are offered a quantitative
dissimilar sources using the Decision-Making analysis based on statistical methodology rooted in the Dempster-Shafer
Dempster-Shafer theory.” theory of evidence that can enhance their decision making processes
regarding the identity of detected objects.
Caselton, W. F. “Decision-Making with Water Resources Research Decision-Making, A water resources example of an application of the Dempster-Shafer
and W. B. Luo Imprecise Probabilities : 28(12): 3071-3083 (1992). Water Resources approach is presented, and the results contrasted with those obtained
Dempster-Shafer Theory Management from the closest equivalent Bayesian scheme.
Chang, Y. C., J. “Evidential reasoning for Civil Engineering Systems Decision-Making, This research proposes a formal methodology for integrating subjective
R. Wright, et al. assessing environmental 14(1): 55-77 (1996). inferential reasoning and geographic information systems (GIS) into a
impact.” decision support system for use in these problem domains. The rationale
for inferential spatial models, and the structure and function of a spatial
modeling environment based on the Dempster-Shafer theory of evidence
are presented.

Class, F., A. “Soft-decision vector (Proceedings ICASSP, Speech The authors describe an algorithm for soft-decision vector quantization
Kaltenmeier, et quantization based on the
IEEE International Recognition, (SVQ) implemented in the acoustic front-end of a large-vocabulary
al. Dempster/Shafer theory.”Conference on Acoustics, Decision-Making speech recognizer based on discrete density HMMs (hidden Markov
Speech and Signal models) of small phonetic units.
Processing 1: 665-668,
(1991).
deKorvin, A., S. “Evaluating policies based Stochastic Analysis and Decision-Making, We use the Dempster-Shafer theory together with techniques of Norton
Hashemi, et al. on their long term average Applications 18(6): 901- Policy selection and Smets to approximate the transition probabilities for an application
cost.” 919 (2000). in policy selection from a set of possible policies in the long term.

Appendix A-8
 DECISION-MAKING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Dekorvin, A. and “A Dempster-Shafer-Based IEEE Transactions on Decision-Making, The Dempster-Shafer theory of evidence is applied to the technology
M. F. Shipley Approach to Compromise Engineering Management Product Selection assessment problem of selecting computer software
Decision-Making with 40(1): 60-67 (1993).
Multiattributes Applied to
Product Selection.”

Denoeux, T. “Modeling vague beliefs Fuzzy Sets and Systems Decision-Making, We introduce the concepts of interval-valued and fuzzy-valued belief
using fuzzy-valued belief 116(2): 167-199 (2000). Classification structures and discuss the application of this framework in the areas of
structures.” decision making under uncertainty and classification of fuzzy data.

Denoeux, T. and “Induction of decision Proceedings of the IEEE Decision-Making, A new tree-structured classifier based on the Dempster-Shafer theory of
M. S. Bjanger trees from partially International Conference on Classification evidence is presented.
classified data using belief Systems, Man and
functions.” Cybernetics 4: 2923-2928
(2000).
Drakopoulos, E. “Decision Rules for IEEE Transactions on Decision-Making A binary hypothesis testing problem is solved using some simple
and C. C. Lee Distributed Decision Automatic Control 37(1): 5- concepts of Dempster-Shafer theory. Each Decision Maker in a
Networks with 14, (1992). distributed decision networks employs Dempster's combining rule to
Uncertainties.” aggregate its input information for a decision.
Ducey, M. J. “Representing uncertainty Forest Ecology and Decision-Making This paper presents examples of silvicultural decision-making using
in silvicultural decisions: Management 150(3): 199- belief functions for the case of no data, sparse data, and adaptive
an application of the 211 (2001). management under increasing data availability.
Dempster-Shafer theory of
evidence.”

Engemann, K. J., “Decision making with International Journal of Decision-Making, We then propose a methodology for decision making under uncertainty,
H. E. Miller, et belief structures: An Uncertainty Fuzziness and Risk Management integrating the ordered weighted averaging aggregation operators and the
al. application in risk Knowledge-Based Systems Dempster-Shafer belief structure. The proposed methodology is applied
management.” 4(1): 1-25 (1996). to a real world case involving risk management at one of the nation's
largest banks.

Appendix A-9
 DECISION-MAKING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Gaglio, S., R. “On the Acquisition and Decision-Making Problems relevant to the construction of a rule-based decision-support
Minciardi, et al. Processing of Uncertain system that is based on uncertain knowledge are addressed. The
Information in Rule-Based representation of uncertainty and the combination of evidence are carried
Decision Support out by means of the Dempster-Shafer theory of evidence.
Systems.”

Garribba, S. F. “Evidence Aggregation in Lecture Notes in Computer Expert Judgments,

and A. Servida Expert Judgments.” Science 313: 385-400, Decision-Making
(1988).

Kohlas, J. and P.- “Theory of Evidence--A Mathematical Methods of Decision Analysis, Evidence theory has been used to represent uncertainty in expert systems,
A. Monney Survey of its Mathematical Operations Research 39: 35- Statistical Analysis, especially in the domain of diagnostics. It can be applied to decision
Foundations, Applications 68, (1994). Imaging, Project analysis and it gives a new perspective for statistical analysis. Among its
and Computational Planning, further applications are image processing, project planning and
Aspects.” Scheduling, Risk scheduling and risk analysis. The computational problems of evidence
Analysis theory are well understood and even though the problem is complex,
efficient methods are available.

Shipley, M. F., C. “Project management: Annual Conference of the Decision-Making, Fuzzy logic and the Dempster-Shafer theory of evidence is applied to an
A. Dykman, et al. Using fuzzy logic and the North American Fuzzy Project IS multiattribute decision making problem whereby the project manager
Dempster-Shafer theory of Information Processing Management must select project team members from candidates, none of whom may
evidence to select team Society NAFIPS: 640-644 exactly satisfy the ideal level of skills needed at any point in time.
members for the project (1999).
duration.”

Smets, P. “The transferable belief Analysis and Management Decision-Making We show how the transferable belief model can be used to assess and to
model for expert of Uncertainty: Theory and combine expert opinions. The transferable belief model has the
judgements.” Applications. B. M. Ayyub, advantage that it can handle weighted opinions and their aggregation
M. M. Gupta and L. N. without the introduction of any ad hoc methods.
Kanal. New York, North-
Holland. 13: 165-170
(1992).

Appendix A-10
 DECISION-MAKING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Srivastava, R. P. “Evidential reasoning for Proceedings of the Hawaii Decision-Making In this paper we develop an evidential network model for `WebTrust
and T. J. Mock WebTrust Assurance International Conference on Assurance,' a service recently proposed by the American Institute of
services.” System Sciences 170: 170 Certified Public Accountants and the Canadian Institute of Chartered
(1999). Accountants. The aggregation of evidence and the resolution of
uncertainties in the model follow the approach of Dempster-Shafer
theory of belief functions.

Yang, J. B. and “An Evidential Reasoning IEEE Transactions on Decision-Making A new evidential reasoning based approach is proposed that may be used
M. G. Singh Approach for Multiple- Systems Man and to deal with uncertain decision knowledge in multiple-attribute decision
Attribute Decision-Making Cybernetics 24(1): 1-18 making (MADM) problems with both quantitative and qualitative
with Uncertainty.” (1994). attributes. This approach is based on an evaluation analysis model and
the evidence combination rule of the Dempster-Shafer theory.

ENGINEERING AND OPTIMIZATION

AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Alim, S. “Application of Dempster- Journal of Structural
Shafer Theory for Engineering-Asce 114(9):
Interpretation of Seismic 2070-2084 (1988).
Parameters.”

Butler, A. C., F. “Computer-Aided-Design AI Edam-Artificial Computer Aided Research in computer-aided design/engineering (CAD/E) has focused on
Sadeghi, et al. Engineering of Bearing Intelligence for Engineering Design, enhancing the capability of computer systems in a design environment,
Systems Using the Design Analysis and Engineering, and this work has continued in this trend by illustrating the use of the
Dempster-Shafer Theory.” Manufacturing 9(1): 1-11, Expert System, Dempster-Shafer theory to expand the computer's role in a CAD/E
(1995). Selection environment. An expert system was created using Dempster-Shafer
methods that effectively modeled the professional judgment of a skilled
tribologist in the selection of rolling element bearings.

Appendix A-11
 ENGINEERING AND OPTIMIZATION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Chen, L. and S. “A modified Dempster- Engineering Optimization Engineering, A new methodology, based on a modified Dempster-Shafer (DS) theory,
S. Rao Shafer theory for 30(3-4): 177-201, (1998). Multicriteria is proposed for solving multicriteria design optimization problems. The
multicriteria optimization.” Design design of a mechanism in the presence of seven design criteria and
Optimization eighteen design variables is considered to illustrate the computational
details of the approach. This work represents the first attempt made in the
literature at applying DS theory for numerical engineering optimization.

Rao, S. S. and L. “Generalized hybrid AIAA Journal 34(8): 1709- Optimization, A generalized hybrid approach is presented for the multiobjective
Chen method for fuzzy 1717, (1996). Engineering optimization of engineering systems in the presence of objectives and
multiobjective Systems constraints that are partly fuzzy and partly crisp. The methodology is
optimization of based on both fuzzy-set and Dempster-Shafer theories to capture the
engineering systems.” features of incomplete, imprecise, uncertain, or vague information that is
often present in real-world engineering systems. The original partly fuzzy
multiobjective optimization problem is first defuzzified into a crisp
generalized multiobjective optimization problem using fuzzy-set theory.
The resulting multiobjective problem is then transformed into an
equivalent single-objective optimization problem using a modified
Dempster-Shafer theory. The computational details of the approach are
illustrated with a structural design example.

Yang, J. B. and “Multiple attribute design Journal of Engineering Evaluation, This paper reports the application of an evidential reasoning approach to
P. Sen evaluation of complex Design 8(3): 211-230, Engineering design selection of retro-fit options for complex engineering products.
engineering products using (1997). Product Selection, The particular selection problem investigated in this paper is initially
the evidential reasoning Decision-Making modeled by means of techno-economic analysis and may be viewed as a
approach.” multiple-attribute decision-making problem with a hierarchical structure
of attributes which may be measured for each design option using
numerical values or subjective judgments with uncertainty.

Appendix A-12
 EXPERT SYSTEMS
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Berenji, H. and “Application of Plausible Proc Am Control Conf: Expert Systems, The authors discuss techniques used for development of knowledge-
H. Lum, Jr. Reasoning to AI-Based 1655-1661. (1987). Knowledge-Based based (e. g. , expert) systems. Specifically, the MYCIN expert system
Control Systems.” Systems certainty factor approach, probabilistic approach, Dempster-Shafer
theory of evidence (1976), possibility theory and linguistic variables, and
fuzzy control are presented.

Ferrier, G. and G. “An integrated GIS and International Journal of Knowledge-base, Approximate reasoning techniques to handle the vagueness and
Wadge knowledge-based system as Geographical Information Geography uncertainty inherent in a large amount of geological data, knowledge and
an aid for the geological Science 11(3): 281-297, reasoning are reviewed with particular emphasis on provenance analysis
analysis of sedimentary (1997). using subjective probability theory, Dempster-Shafer theory and fuzzy
basins.” logic techniques.

Gammerman, A., “Sysex: An Expert System Proceedings of SPIE The Identification, The aim of this research is to create an expert system which would help
B. Skullerud, et for Biological International Society for Biology, Expert with the task of identifying a biological specimen. The Dempster-Shafer
al. Identification.” Optical Engineering 657: Systems theory of evidence was used to handle uncertainty associated with the
34-39, (1986). date and the expertise.
Guan, J., D. A. “Dempster-Shafer theory (1990). IEE Colloquium: Expert Systems, Dempter-Shafer theory is discussed, focusing on the union of the
Bell, et al. and rule strengths in expert 086. Fault Diagnosis granules in a granule set. The discussion is illustrated by considering an
systems.” example of fault diagnosis in a distributed vehicle monitoring system.

Shenoy, P. P. “Using Dempster-Shafer's Proceedings of SPIE The Expert Systems, The main objective of this paper is to describe how Dempster-Shafer's
belief-function theory in International Society for Valuation-Based (DS) theory of belief functions fits in the framework of valuation-based
expert systems.” Optical Engineering: 2-14, Systems systems (VBS). Since VBS serves as a framework for managing
(1992). uncertainty in expert systems, this facilitates the use of DS belief-
function theory in expert systems.

Stephanou, H. E. “Evidential Framework for Control Systems, The author deals with a class of knowledge-based control systems that
Intelligent Control.” Decision-Making involve two types of (not necessarily probabilistic) uncertainty: (1) an
incomplete set of control rules contributed by multiple domain experts;
and (2) incomplete and/or inaccurate feedback information from multiple
sensors. The Dempster-Shafer theory of evidence provides the basic
framework for the representation of uncertain knowledge.

Appendix A-13
 EXPERT SYSTEMS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Wunsch, G. and “Expert systems to assist Journal Fur Praktische Expert Systems, Certain, uncertain and lacking knowledge has to be considered for
F. Klages analytical chemistry: Chemie-Chemiker-Zeitung Analytical intelligent counseling. In the ICP mass spectrometry the composition of
Realization of learning 338(7): 593-597, (1996). Chemistry, Mass the actual sample and the ionization rates are the most important
ability and plausibility Spectrometry, parameters to be prognosticated. The way of storage and retrieval of data
checking demonstrated for Decision-Making and of decision making should be automatically checked and improved
ICP mass spectrometry.” with respect to the success of previous guesses. The Dempster-Shafer
theory is used for the combination and propagation of uncertainties.

FAULT DIAGNOSIS AND FAILURE DETECTION

AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Kang, H., J. “An application of fuzzy Proceedings of the IEEE Failure Detection, A novel approach to failure detection and identification (FDI) is
Cheng, et al. logic and Dempster-Shafer Conference on Decision and Identification, proposed which combines an analytic estimation method and an
theory to failure detection Control 2: 1555-1560, Decision-Making intelligent identification scheme in such a way that sensitivity to true
and identification.” (1991). failure modes is enhanced, while the possibility of false alarms is
reduced. At the final stage of the algorithm, an index is computed--the
degree of certainty--based on Dempster-Shafer theory, which measures
the reliability of the decision. The FDI algorithm has been applied
successfully to the detection of rotating stall and surge instabilities in
axial flow compressors.

Parikh, C. R., M. “Application of Dempster- Pattern Recognition Letters Classification, Fault This paper is concerned with the use of Dempster-Shafer theory in
J. Pont, et al. Shafer theory in condition 22(6-7): 777-785, (2001). Diagnosis, 'fusion' classifiers. We demonstrate the effectiveness of this approach in
monitoring applications: a Monitoring a case study involving the detection of static thermostatic valve faults in
case study.” a diesel engine cooling system.

Appendix A-14
 FAULT DIAGNOSIS AND FAILURE DETECTION (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Tanaka, K. and “Design condition for Reliability Engineering and Safety Monitoring, The present article proposes two types of an automatic monitoring
G. J. Klir incorporating human System Safety 65: 251-258, Sensors, Failure system not involving any human inspection or a human-machine (H-M)
judgement into monitoring (1999). Detection cooperative monitoring system with inspection. In order to compare the
systems.” systems, an approach based on the Dempster-Shafer theory is proposed
for uncertainty analysis. By comparing their expected losses as a result of
failed dangerous failures or failed safe failures as well as the inspection
errors, the condition is determined under which H-M cooperative systems
incorporating human judgements are more effective than automatic
monitoring systems.

Vachtsevanos G., “Detection and Journal of Guidance Identification, A new approach to failure detection and identification is proposed that
H. Kang, et al. Identification of Axial- Control and Dynamics Failure Detection combines an analytic estimation method and an intelligent identification
Flow Compressor 15(5): 1216-1223, (1992). scheme in such a way that sensitivity to true failure modes is enhanced
Instabilities.” while the possibility of false alarms is reduced. We employ a real-time
recursive parameter estimation algorithm with covariance resetting that
triggers the fault detection and identification routine only when potential
failure modes are anticipated. A possibilistic scheme based on fuzzy set
theory is applied to the identification part of the algorithm with
computational efficiency. At the final stage of the algorithm, an index is
computed-the degree of certainty-based on Dempster-Shafer theory,
which measures the reliability of the decision. The proposed algorithm
has been applied successfully to the detection of rotating stall and surge
instabilities in axial flow compressors.

Van Dam, K. and “Extension of Dempster- IEE Conference Fault Diagnosis, A novel method is presented for propagating uncertainty that also
T. J. Moulsley Shafer theory and Publication(395): 310-315, Communication calculates measures of contradictions in the input data. This method can
application to fault (1994). Systems improve the performance of a Reason Maintenance System (RMS) by
diagnosis in ranking the contradictions and resolving the most severe of these first.
communication systems.” An example shows the application of this technique to fault diagnosis in
a communication system.

Appendix A-15
 IMAGE PROCESSING
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Askari, F. and B. “Neural network Oceans Conference Record Image Processing, This paper discusses an approach for automatic feature detection and
Zerr architecture for automatic 2: 1017-1021 (1998). Oceanography, sensor fusion in remote sensing imagery using a combination of neural
extraction of Satellite, Neural network architecture and Dempster-Shafer theory of evidence.
oceanographic features in Networks
satellite remote sensing
imagery.”

Aslandogan, Y. “Evaluating strategies and Proceedings of the ACM Image Processing, We provide experimental evaluation of the following strategies for the
A. and C. T. Yu systems for content based International Multimedia Multimedia, content based indexing of multimedia: i) Face detection on the image
indexing of person images Conference and Exhibition: Recognition, followed by Text/HTML analysis of the containing page; ii) face
on the web.” 313-321 (2000). Content-based detection followed by face recognition; iii) face detection followed by a
Indexing linear combination of evidences due to text/HTML analysis and face
recognition; and iv) face detection followed by a Dempster-Shafer
combination of evidences due to text/HTML analysis and face
recognition.
Betz, J. W., J. L. “Representation and Image Processing, A framework is presented for deriving and transforming evidence-
Prince, et al. transformation of Artificial Vision, theoretic belief representations of uncertain variables that denote
uncertainty in an evidence Sensors numerical quantities. Belief is derived from probabilistic models using
theory framework.” relationships between probability bounds and the support and plausibility
functions used in evidence theory. This model-based approach to belief
representation is illustrated by an algorithm currently used in a vision
system to label anomalous high-intensity pixels in imagery.

Bloch, I. “Some aspects of Pattern Recognition Letters Image Processing, This paper points out some key features of Dempster-Shafer evidence
Dempster-Shafer evidence 17(8): 905-919 (1996). Medicine, theory for data fusion in medical imaging. Examples are provided to
theory for classification of Classification, show its ability to take into account a large variety of situations, which
multi-modality medical Medical Imaging actually often occur and are not always well managed by classical
images taking partial approaches nor by previous applications of Dempster-Shafer theory in
volume effect into medical imaging.
account.”

Appendix A-16
 IMAGE PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Chapron, M. “A color edge detector Image Processing, Segmentation based on contour detection is a relevant stage before image
based on Dempster-Shafer Pattern Recognition interpretation or pattern recognition. This paper is concerned with color
theory.” image filtering and color edge detecting. These 2 techniques utilize the
Dempster-Shafer theory.
Huber, R. “Scene classification of Image and Vision Image Processing, This paper presents a method for fusion of information derived from
SAR images acquired from Computing 19(13): 1001- Radar, different airborne synthetic aperture radar measurement processes and
antiparallel tracks using 1010 (2001). Classification from different observations of the same scene. Dempster-Shafer theory
evidential and rule-based will be used to fuse radar backscatter and phase measurements.
fusion.”

Ip, H. H. S. and “Evidential reasoning for Australian and New Image Processing, The Dempster-Shafer theory of evidential reasoning is applied to
R. C. K. Chiu facial gestures recognition Zealand Conference on Recognition combine evidence represented by the facial features. The study
from cartoon images.” Intelligent Information demonstrates the feasibility of applying the Dempster-Shafer theory to
Systems Proceedings: 397- facial gesture recognition.
401 (1994).
Ip, H. H. S. and J. “Human face recognition IEEE International Image Processing, In this paper, image processing techniques developed for the extraction
M. C. Ng using Dempster-Shafer Conference on Image Recognition of the set of visual evidence, the formulation of the face recognition
theory.” Processing 1: 292-295 problem within the framework of Dempster-Shafer Theory and the
(1994). design of suitable mass functions for belief assignment are discussed.

Janez, F., O. “Automatic map updating Proceedings of SPIE The Image Processing, In this article, we present a strategy to report in an automatic way
Goretta, et al. by fusion of multispectral International Society for Cartography significant changes on a map by fusion of recent images in various
images in the Dempster- Optical Engineering 4115: spectral bands. For configurations of partial overlapping between map
Shafer framework.” 245-255 (2000). and images, it is difficult or even impossible to formalize the approach
suggested within a probabilistic framework. Thus, the Dempster-Shafer
theory is shown as a more suitable formalism in view of the available
information, and we present several solutions.

Appendix A-17
 IMAGE PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Kasinski, A. and “Managing processes of Proceedings of the IEEE Image Processing, Three lower level layers of the hierarchical machine perception system
T. Piascik perceptual organization for International Conference on Artificial Vision are described, and experimental results are provided. Three approaches
emerging geometrical Systems, Man and to the features fusion: one based on crisp, geometrical, heuristic
objects.” Cybernetics 3: 1604-1609 conditions, the second based on fuzzyfied conditions and the third one
(2000). based on Dempster-Shafer theory are addressed. The results obtained
with three methods are compared and presented on the example image of
the real scene.

Krishnapuram, R. “A Belief Maintenance International Journal of Image Processing, In this article, we show how the Dempster-Shafer theoretic concepts of
Scheme for Hierarchical Intelligent Systems 6(7): Knowledge-base, refinement and coarsening can be used to aggregate and propagate
Knowledge-Based Image- 699-715 (1991). Decision-Making evidence in a multi-resolution image analysis system based on a
Analysis Systems.” hierarchical knowledge base.

Lohmann, G. “Evidential reasoning Forschungsbericht Deutsche Image Processing, A new algorithm for classifying satellite images is presented. The new
approach to the Forschungsanstalt fuer Luft Satellite, algorithm called EBIS (Evidence-Based Interpretation of Satellite
classification of satellite und Raumfahrt, DLR FB: Classification, Images) - will be used for ecological mappings. In EBIS, a feature space
images.” 91-29 (1991). Ecological is regarded as a source of evidence in the sense of the Dempster-Shafer-
Mapping theory, and methods of evidential reasoning are used for combining
evidence stemming from several disparate sources. This makes EBIS
particularly useful for integrating different data sources such as various
sensors, digital elevation models or other types of ancillary data.

Mulhem, P., D. “Labeling update of IEEE International Image Processing, We propose here to use conceptual graphs (a knowledge representation
Hong, et al. segmented images using Conference on Multi Media Object Recognition formalism that allow fast processing) with Dempster-Shafer theory of
conceptual graphs and and Expo(II): 1129-1132 evidence to update original labeling coming from a segmentation that
Dempster-Shafer theory of (2000). labels image regions out of context.
evidence.”

Appendix A-18
 IMAGE PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Payne, M. G., Q. “Using the Dempster- Proceedings of SPIE The Image Processing We present an algorithm that performs pixel-level segmentation based
Zhu, et al. Shafer reasoning model to International Society for upon the Dempster-Shafer theory of evidence. The algorithm fuses image
perform pixel-level Optical Engineering: 26-35 data from the multichannels of color spectra. Dempster-Shafer reasoning
segmentation on color (1992). is used to drive the evidence accumulation process for pixel level
images.” segmentation of color scenes.

Peddle, D. R. “Mercury-Circle-Plus : An (1995). Computers & Image Processing, MERCURY circle plus is a multisource evidential reasoning
Evidential Reasoning Geosciences 21(10): 1163- Software, classification software system based on the Dempster-Shafer theory of
Image Classifier.” Geoscience, evidence. The design and implementation of this software package is
Environment, described for improving the classification and analysis of multisource
Classification digital image data necessary for addressing advanced environmental and
geoscience applications. An example of classifying alpineland cover and
permafrost active layer depth in northern Canada is presented to illustrate
the use and application of these ideas.

Pinz, A., M. “Active fusion : A new Pattern Recognition Letters Image Processing, In this paper, we introduce a new method, termed ''active fusion'', which
Prantl, et al. method applied to remote 17(13): 1349-1359 (1996). Artificial Vision provides a common framework for active selection and combination of
sensing image information from multiple sources in order to arrive at a reliable result at
interpretation.” reasonable costs. The implementation of active fusion on the basis of
probability theory, the Dempster-Shafer theory of evidence and fuzzy
sets is discussed.

Sarma, L. C. S. “A Prototype Expert- Sadhana-Academy Image Processing, In this paper, we have critically studied visual interpretation processes
and V. V. S. System for Interpretation Proceedings in Engineering Knowledge-base, for urban land cover and land use information. The Dempster-Shafer
Sarma of Remote-Sensing Image Sciences 19(pt.1)): 93-111 Classification, theory of evidence is used to combine evidence from various
Data.” (1994). Geography interpretation keys for identification of generic class and subclass of a
logical image object. Analysis of some Indian Remote Sensing Satellite
images has been done using various basic probability assignments in
combination with learning.

Appendix A-19
 IMAGE PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Tupin, F., I. “A first step toward IEEE Transactions on Image Processing, We propose a method aiming to characterize the spatial organization of
Bloch, et al. automatic interpretation of Geoscience and Remote Radar, Cartography the main cartographic elements of a synthetic aperture radar (SAR)
SAR images using Sensing 37(3/pt.1): 1327- image and thus giving an almost automatic interpretation of the scene.
evidential fusion of several 1343 (1999). Our approach is divided into three main steps which build the whole
structure detectors.” image interpretation gradually. The first step consists of applying low-
level detectors taking the speckle statistics into account and extracting
some raw information from the scene. The detector responses are then
fused in a second step using Dempster-Shafer theory, thus allowing the
modeling of the knowledge that we have about operators, including
possible ignorance and their limits. A third step gives the final image
interpretation using contextual knowledge between the different classes,
Results of the whole method applied to different SAR images and to
various landscapes are presented.

Vancleynen- “Road Extraction from Pattern Recognition Letters Image Processing, Road networks extracted from multi-temporal SPOT images of the same
breugel, J., S. A. Multitemporal Satellite 12(6): 371-380 (1991). Satellite scene are matched to collect evidence for individual road segments. The
Osinga, et al. Images by an Evidential Dempster-Shafer theory is applied to find a degree of confirmation for a
Reasoning Approach.” road segment in one network based on its corresponding lines in the
other networks.
Vannooren- “Color image segmentation IEEE International Image Processing, In this paper, we propose a color image segmentation method based on
berghe, P., O. using Dempster-Shafer's Conference on Image Biomedical the Dempster-Shafer's theory. The basic idea consists in modeling the
Colot, et al. theory.” Processing 4: 300-304 Engineering color information in order to have the features of each region in the
(1999). image. This model, obtained on training sets extracted from the intensity,
allows for the reduction of the classification errors concerning each pixel
of the image. The proposed segmentation algorithm has been applied to
synthetic and biomedical images in order to illustrate the methodology.

Verly, J. G., R. L. “Model-Based System for Optical Engineering 31(12): Image Processing, We describe an experimental model-based automatic target recognition
Delanoy, et al. Automatic Target 2540-2552 (1992). Target Recognition, (ATR) system, called XTRS, for recognizing 3-D vehicles in real or
Recognition from Forward- Radar synthetic, ground-based or airborne, 2-D laser-radar range and intensity
Looking Laser-Radar images
Imagery.”

Appendix A-20
 IMAGE PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Verly, J. G., B. “Automatic Object Image Processing, A system for the automatic recognition of objects in real infrared-radar
D. Williams, et Recognition from Range Radar, Object range imagery is described. Recognition consists of matching symbolic
al. Imagery Using Appearance Recognition range silhouette descriptions against appearance models of known
Models.” objects and then deciding among the possible objects using the Dempster-
Shafer theory of evidence. Both contour-based and region-based
silhouette extraction and recognition showed good results.

Wang, Y. and D. “Evidential reasoning- Canadian Journal of Image Processing, In this paper, a two-stage distribution-free classification strategy was
L. Civco based classification of Remote Sensing 20: 381- Classification, adopted to incorporate ancillary data in remote sensing image
multi-source spatial data 395 (1994). Cartography classification.. The approach provides a scheme that can readily pool
for improved land cover attribute information from multi-source spatial data.
mapping.”
Wilkinson, G. G. “Evidential reasoning in a International Journal of Image Processing, This paper presents a technique for integrating diverse sources of
and J. Megier pixel classification Remote Sensing 11: 1963- Classification, evidence about pixel or segment classification using the belief function
hierarchy. A potential 1968, (1990). Expert System, approach of the Dempster-Shafer theory of evidential reasoning. A
method for integrating Geography description of the algorithm is provided and a case study is given to
image classifiers and illustrate the potential of the method for combining output from image
expert system rules based classifers, geographic information systems and expert system rules
on geographic context.” concerning geographic context.

Yamane, S., K. “Model-Based Object Systems and Computers in Image Processing, A method of object recognition is proposed based on Dempster-Shafer
Aoki, et al. Recognition Using Basic Japan 26(12): 49-57 (1995). Object Recognition theory (DS theory), which can treat the ambiguity of image data.
Probability Assignment.”
MEDICAL APPLICATIONS
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Andress, K. M. “Evidential reconstruction IEEE International Medicine, Digital The Dempster-Shafer theory of evidence is used to combine information
of vessel trees from Conference on Image Imagery about location of the vessels from the different projections contained in
rotational angiograms.” Processing 3: 385-389, the digital subtraction angiogram (DSA) sequence.
(1998).

Appendix A-21
 MEDICAL APPLICATIONS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Bell, D. A., J. W. “Using the Dempster- Kybernetes 27(4-5): 511-&, Spatial Reasoning The objects of interest here are geometric forms, and we can encode
Guan, et al. Shafer orthogonal sum for (1998). rectangular and other shaped forms using hexadecimal numbers
reasoning which involves according to shapes and positions. Boolean algebra of such shapes can
space.” then be used directly in Dempster-Shafer-type reasoning exercised.
Discusses how medical and other fields can gain from this approach.

Boston, J. R., L. “Combination of data IEEE Conference on Control, Detection, Data fusion techniques for ventricular suction detection in a heart assist
Baloa, et al. approaches to heuristic Control Applications Medicine device based on Bayesian, fuzzy logic and Dempster-Shafer theory were
control and fault Proceedings 1: 98-103, evaluated. Fusion techniques based on fuzzy logic and Dempster-Shafer
detection.” (2000). theory provide a measure of uncertainty in the fused result. This
uncertainty measure can be used in the control process, and it can also be
used to identify faults in pump operation.
Cios, K. J., R. E. “An Expert System for Computer Applications in
Freasier, et al. Diagnosis of Coronary- the Biosciences 6(4): 333-
Artery Stenosis Based on 342, (1990).
Ti-201 Scintigrams Using
the Dempster-Shafer
Theory of Evidence.”

Lefevre, E., O. “Knowledge modeling Proceedings of the IEEE Medical Imaging The aim of this paper is to present modeling methods of knowledge for
Colot, et al. methods in the framework International Conference on the initialization of belief functions from Dempster-Shafer theory.
of evidence theory. An Systems, Man and Moreover, an experimental comparison of these different modeling on
experimental comparison Cybernetics 4: 2806-2811, real data extracted from images of dermatological lesions is presented.
for melanoma detection.” (2000).

Medina, R., M. “Evidence combination Medicine, Image A left ventricle three-dimensional reconstruction method from two
Garreau, et al. approach to reconstruction orthogonal angiographic projections is described based on the cylindrical
of the left ventricle from closure operation and the Dempster-Shafer Theory.
two angiographic views.”

Appendix A-22
 MEDICAL APPLICATIONS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Principe, J. C., S. “Sleep staging automaton IEEE Transactions on Medicine, The Dempster-Shafer theory of evidence is used to develop a model for
K. Gala, et al. based on the theory of Biomedical Engineering 36: Biomedical automated sleep staging by combining signal information and human
evidence.” 503-509(1989). Engineering, Signal heuristic knowledge in the form of rules.
Information

Suh, D. Y., R. L. “Knowledge-based system IEEE Transactions on Medicine, A strategy for a knowledge-based system to detect the interior and
Eisner, et al.. for boundary detection of Medical Imaging 12(1): 65- Knowledge-based exterior boundaries of the left ventricle from time-varying cross-sectional
four-dimensional cardiac 72 (1993). system, Image, images obtained by ECG-gated magnetic resonance imaging uses both
magnetic resonance image Multiple sources fuzzy set theory and Dempster and Shafer theory to manage the
sequences.” knowledge and to control the flow of system information.

Suh, D. Y., R. M. “A system for knowledge- Proceedings ICASSP, IEEE Medicine, A knowledge-based system is described for boundary detection from
Mersereau, et al. based boundary detection International Conference on Knowledge-based magnetic resonance image sequences of a beating heart. It is shown that
of cardiac magnetic Acoustics, Speech and system, Image, the Dempster/Shafer theory and fuzzy set theory can be used for control
resonance image Signal Processing 4: 2341- Boundary of the system as well as for labeling objects in the images.
sequences.” 2344 (1990). Detection

MISCELLANEOUS
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Cai, D., M. F. “Knowledge discovery in International Journal of Knowledge Distributed databases allow us to integrate data from different sources
McTear, et al. distributed databases using Intelligent Systems 15(8): Discovery, which have not previously been combined. Evidential functions are
evidence theory.” 745-761 (2000). Distributed suited to represent evidence from different sources. Previous work has
Databases defined linguistic summaries to discover knowledge by using fuzzy set
theory and using evidence theory to define summaries. In this paper we
study linguistic summaries and their applications to knowledge discovery
in distributed databases.

Appendix A-23
 MISCELLANEOUS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Cortes Rello, E. “Uncertain reasoning using Expert Systems 7: 9-18 Forecasting, The intended purpose of this article is twofold: to study techniques for
and F. Golshani the Dempster-Shafer (1990). Marketing, uncertainty management in expert systems, particularly the Dempster-
method. An application in Management, Shafer theory of belief; and to use this method in the construction of an
forecasting and marketing Expert Systems expert system for the field of forecasting and marketing management.
management.”

Gillett, P. R. “Monetary unit sampling: a International Journal of Finance Audit procedures may be planned and audit evidence evaluated using
belief-function Approximate Reasoning Applications monetary unit sampling (MUS) techniques within the context of the
implementation for audit 25(1): 43-70, (2000). Dempster-Shafer theory of belief functions.
and accounting
applications.”

Golshani, F., E. “Dynamic route planning Knowledge Based Systems Autonomous The paper describes the design of a route planning system, called RUTA-
Cortes Rello, et with uncertain 9: 223-232, (1996). Vehicle Navigation 100, that works with incomplete information obtained from many
al. information.” unreliable knowledge sources and plans an optimal route by minimizing
both danger and distance. The Dempster-Shafer theory of belief is used
as the underlying formalism to pool and represent uncertain information
and reason with it.

Isaksen, G. H. “Interpretation of Organic Geochemistry 26(1- Molecular This paper describes the application of fuzzy logic and Dempster-Shafer
and C. S. Kim molecular geochemistry 2): 1-10, (1997). Geochemistry theory to the interpretation of molecular geochemistry data with respect
data by the application of to key exploration parameters, such as thermal maturity, organic facies
artificial intelligence (organic matter type and depositional environment of the source rock(s)),
technology.” geological age, and the degree of biodegradation.

Ji, Q., M. M. “Evidential reasoning Manufacturing, This paper introduces an evidential reasoning based approach for
Marefat, et al. approach for recognizing Feature Extraction recognizing and extracting manufacturing features from solid model
shape features.” description of objects. The main contributions of our approach include
introducing the evidential reasoning (Dempster-Shafer theory) to the
feature extraction domain and developing the theory of principle of
association to overcome the mutual exclusiveness assumption of the
Dempster-Shafer theory.

Appendix A-24
 MISCELLANEOUS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Ji, Q., M. M. “Dempster-Shafer and Proceedings of the National Manufacturing, The paper evaluates the performance the Dempster-Shafer theory (DS)
Marefat, et al. Bayesian networks for Conference on Artificial Feature Extraction and the Bayesian Belief Network (BBN) with regard to their ability to
CAD-based feature Intelligence 2, (1994). extract manufacturing features from the solid model description of
extraction: A comparative objects.
investigation and analysis.”

Kawahara, A. A. “An Application of Energy 17(3): 205-214, Biogas Technology We apply the Dempster-Shafer theory of belief functions to the
and P. M. Dempster-Shafer Theory to (1992). assessment of biogas technology in rural areas of Brazil. Two case
Williams the Assessment of Biogas studies are discussed in detail and the results compared with a more
Technology.” conventional method of project appraisal. On the computational side, it is
shown how local computation and dimensionality reduction, in cases
where certain relations hold between variables, can increase efficiency.

Lalmas, M. “Dempster-Shafer's theory SIGIR Forum: 110-118, Document Chiaramella et al advanced a model for indexing and retrieving
of evidence applied to (1997). Retrieval, structured documents. This paper adds to this model a theory of
structured documents: Document uncertainty, the Dempster-Shafer theory of evidence. It is shown that the
Modelling uncertainty.” Structure, theory provides a rule, the Dempster's combination rule, that allows the
Information expression of the uncertainty with respect to parts of a document, and
Retrieval, Indexing that is compatible with the logical model developed by Chiaramella et al.

Lalmas, M. and I. “Representing and Journal of Documentation Document In this paper we report on a theoretical model of structured document
Ruthven retrieving structured 54(5): 529-565, (1998). Retrieval, indexing and retrieval based on the Dempster-Shafer Theory of
documents using the Document Evidence. This includes a description of our model of structured
Dempster-Shafer theory of Structure, document retrieval, the representation of structured documents, the
evidence: Modelling and Information representation of individual components, how components are combined,
evaluation.” Retrieval, Indexing details of the combination process, and how relevance is captured within
the model.

Appendix A-25
 MISCELLANEOUS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Laskey, K. B. and “Applications of the Winter Simul Conf Proc: Simulation The key feature of the Dempster-Shafer theory is that precision in inputs
M. S. Cohen Dempster-Shafer Theory of 440-444, (1986). is required only to a degree justified by available evidence. The output
Evidence for Simulation.” belief function contains an explicit measure of the firmness of output
probabilities. The authors give an overview of belief function theory,
presents the basic methodology for application to simulation, and gives a
simple example of a simulation involving belief functions.

Ling, X. and W. “Combining opinions from Applied Artificial Expert Opinion, We develop an approach for combining expert opinions that formally
G. Rudd several experts.” Intelligence 3: 439-452, Opinion Pooling allows for stochastic dependence. This approach is based on an extension
(1989). of the Dempster-Shafer theory, a well-known calculus for reasoning with
uncertainty in artificial intelligence.

Luo, W. B. and “Using Dempster-Shafer Journal of Environmental Decision Analysis, This paper presents, along with some elementary examples, aspects of
B. Caselton theory to represent climate Management 49(1): 73-93, Climate Change, the Dempster-Shafer approach that contribute to its appeal when dealing
change uncertainties.” (1997). Water Resource with weak subjective and data-based information sources that have a
Projects bearing on climate change.

Mellouli, K. and “Pooling expert opinions Proceedings of the IEEE Expert Opinion, In this paper, we propose a method based on Dempster-Shafer theory of
Z. Elouedi using Dempster-Shafer International Conference on Opinion Pooling evidence, to pool expert judgements about the hypotheses of the studied
theory of evidence.” Systems, Man and field and to get an assessment and even a ranking of the different
Cybernetics 2: 1900-1905, scenarios.
(1997).

Schocken, S. and “On the use of the International Journal of Information This paper has two objectives: (i) to describe and resolve some caveats in
R. A. Hummel Dempster Shafer model in Man Machine Studies Retrieval, the way the Dempster Shafer theory is applied to information indexing
information indexing and 39(5): 843-879, (1993). Information and retrieval, and (ii) to provide an intuitive interpretation of the
retrieval applications.” Indexing Dempster Shafer theory, as it unfolds in the simple context of a canonical
indexing model.

Appendix A-26
 MISCELLANEOUS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Sy, B. K. and J. “AI-based communication IEEE Transactions on Communication The device is centered on a knowledge base of the grammatical rules and
R. Deller, Jr. system for motor and Biomedical Engineering 36: System, message elements. A belief reasoning scheme based on both the
speech disabled persons: 565-571, (1989). Optimization information from external sources and the embedded knowledge is used
Design methodology and to optimize the process of message search. The search for the message
prototype testing.” elements is conceptualized as a path search in the language graph, and a
special frame architecture is used to construct and to partition the graph.
Bayesian belief reasoning from the Dempster-Shafer theory of evidence
is augmented to cope with time-varying evidence.

MULTIPLE SENSORS
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Basti, Egrave, et “Methods for multisensor Aerospace Science and Multisensor This paper proposes to analyze methods applied to the multisensor
al. classification of airborne Technology 2(6): 401- Classification, classification of airborne targets using Dempster-Shafer theory. Several
targets integrating evidence 411(1998). Airborne Target simulations relating to an airborne target classification problem are
theory.” Classification presented.

Belloir, F., R. “A smart flat-coil eddy- Measurement Science & Multiple Sensors, This paper describes a smart eddy-current sensor for locating and
Huez, et al. current sensor for metal- Technology 11(4): 367-374 Pattern Recognition identifying metal tags used to recognize buried pipes. Intelligent pattern-
tag recognition.” (2000). recognition methods and their combination by the Dempster-Shafer
theory of evidence are briefly presented.

Braun, J. J. “Dempster-Shafer Theory Proceedings of SPIE The Multiple sensors, This paper presents a Monte Carlo simulation approach for a
and Bayesian reasoning in International Society for Classification comparative analysis of a Dempster-Shafer Theory based and a Bayesian
multisensor data fusion.” Optical Engineering 4051: multisensor data fusion in the classification task domain, including the
255-266 (2000). implementation of both formalisms, and the results of the Monte Carlo
experiments of this analysis.

Appendix A-27
 MULTIPLE SENSORS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Coombs, K., D. “Using Dempster-Shafer Proceedings of SPIE The Sensors, Ballistic The Dempster Shafer (DS) Theory of Evidential Reasoning may be
Freel, et al. methods for object International Society for Missile useful in handling issues associated with theater ballistic missile
classification in the theater Optical Engineering 3719: Discrimination discrimination. This paper highlights the Dempster-Shafer theory and
ballistic missile 103-113, (1999). describes how this technique was implemented and applied to data
environment.” collected by two infrared sensors on a recent flight test.

Fabre, S., A. “Sensor Fusion Integrating Multiple Sensors The Dempster-Shafer theory of evidence is used to integrate information
Appriou, et al. Contextual Information.” from the context of the sensor acquisitions.

Jiang, J., J. Guo, “Multisensor multiple- Multiple Sensors, A multisensor multiple-attribute data association method is presented
et al. attribute data association.” Target based on Dempster and Shafer (D-S) evidence theory. This approach is
Identification, illustrated by simulations involving multisensor multiple targets in a
Simulation dense clutter environment.
Pigeon, L., B. “Dempster-Shafer theory Proceedings of SPIE The Sensors, Satellites This study suggests a slight variation of the Dempster-Shafer theory
Solaiman, et al. for multi-satellites International Society for using observation qualification in multi-sensor contexts. The uncertainty
remotely-sensed Optical Engineering 4051: is placed on the rules instead of on sources. Thus, sensor's specialization
observations.” 228-236, (2000). is taken into account. By this approach, the masses are not directly
attributed on the frame of discernment elements, but on the rules
themselves that become the sources of knowledge, in the context of
Dempster combining rule. It proposes then an approach for observation
qualification in a multi-sensor context, as well as it suggests a new path
for the delicate task of mass attribution.

Safranek, R. J., S. “Evidence accumulation IEEE Transactions on Sensors, Artificial Vision sensor output can be processed to yield a multitude of low-level
Gottschlich, et al. using binary frames of Robotics and Automation 6: Vision measurements, where each is inherently uncertain, which must somehow
discernment for 405-417, (1990). be combined to verify the locations of an object. It is shown that this
verification vision.” combination can be accomplished via Dempster-Shafer theory using
binary frames of discernment (BFODs).

Appendix A-28
 MULTIPLE SENSORS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Tang, Y. C. and “A Geometric Feature IEEE Transactions on Identification, The paper presents an effective and reliable procedure for identifying
C. S. G. Lee Relation Graph Systems Man and Sensors, corresponding measurements of features in the presence of sensory
Formulation for Consistent Cybernetics 22(1): 115-129, Optimization, uncertainty based on both geometric and topological constraints, and a
Sensor Fusion.” (1992). Simulation nonlinear programming formulation for maintaining consistency in a
network of relations is proposed. The Dempster-Shafer theory of belief
functions is applied to make the utilization of topological constraints in
achieving reliable identification.

Tang, Y. C. and “Optimal Strategic IEEE Transactions on Object Reliable and knowledge-based recognition of objects is obtained by
C. S. G. Lee Recognition of Objects Systems Man and Recognition, applying the Dempster-Shafer theory of belief functions. Computer
Based on Candidate Cybernetics 22(4): 647-661, Identification, simulations were performed to verify the feasibility and to analyze the
Discriminating Graph with (1992). Optimization, performance of the optimal strategic recognition of objects.
Coordinated Sensors.” Sensors, Simulation

Tchamova, A. “Evidence reasoning Mathematics and Sensors, Data The theory of Dempster-Shafer is discussed with emphasis placed on its
theory with application to Computers in Simulation Association use grown from the field of multisensor data fusion and data association
the identity estimation and 43: 139-142, (1997). Systems, systems. The aims of this paper are to investigate: how the structure of
data association systems.” Simulation multisensor integration systems influences over the accuracy of objects
identification process; to determine the dependence of the degree of
uncertainty on the speed of receiving best evidential intervals; to
determine what is the impact of increasing number of sensors on the
calculation time.

Wang, G., Y. He, “Adaptive sensor Chinese Journal of Multiple Sensors, Sensor management has been an active research area in recent years.
et al. management in multisensor Electronics 8: 136-139 Simulation Based on fuzzy set theory and the Dempster-Shafer theory of
data fusion system.” (1999). mathematical evidence, adaptive sensor management schemes in
multisensor data fusion system are presented by using individual sensor's
performance.
Zhang, R., G. Gu, “AUV obstacle-avoidance Multiple Sensors, This paper presents a method of AUV (Autonomous Underwater
et al. based on information Autonomous Vehicle) obstacle avoidance based on information fusion of multi-
fusion of multi-sensors.” Vehicles sensors. Dempster Shafer's theory of evidence is used to judge whether
an obstacle exists ahead of an AUV.

Appendix A-29
 RISK AND RELIABILITY
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Cronhjort, B. T. “Computer Assisted Risk Control, The authors proceed to suggest an expert systems approach for the
and A. Mustonen Reduction of Vulnerability Expert Systems evaluation of EDP risks, and for risk control. A methodology based on
of Data Centers.” the Dempster-Shafer Theory of Evidence is proposed, and the essential
principles for the implementation of such an expert system are outlined.

Holmberg, J., P. “Application of the Reliability Engineering & Risk Analysis

Silvennoinen, et Dempster-Shafer Theory of System Safety 26(1): 47-58,
al. Evidence for Accident (1989).
Probability Estimates.”
Ibrahim, A. and Uncertainties in risk-based Analysis and Management Risk Analysis, Catastrophic industrial failures over the past decade highlight the societal
B. M. Ayyub inspection of complex of Uncertainty: Theory and Complex Systems need to use more explicitly risk-based methods and procedures with
systems. Applications. B. M. Ayyub, uncertainty analysis for these systems. Three measures of uncertainty are
M. M. Gupta and L. N. discussed and several examples to illustrate their applications are
Kanal. New York, North- presented. Logic diagrams and techniques were utilized to propagate
Holland. 13: 247-262, uncertainties for the process of assessing the magnitude of consequences
(1992). due to failure and the uncertainty associated with them.

Inagaki, T. “Interdependence between IEEE Transactions on Reliability, Safety, This paper explores the application of the Dempster-Shafer theory in
Safety-Control Policy and Reliability 40(2): 182-188, Fault Warning system reliability and safety. Inappropriate application of the Dempster-
Multiple-Sensor Schemes (1991). Shafer theory to safety-control policies can degrade plant safety. This is
Via Dempster-Shafer proven in two phases: 1) A new unified combination rule for fusing
Theory.” information on plant states given by independent knowledge sources such
as sensors or human operators is developed. 2) Combination rules can
not be chosen in an arbitrary manner; ie, the best choice of combination
rules depends on whether the safety-control policy is fault-warning or
safety-preservation.

Appendix A-30
 ROBOTICS
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Clerentin, A., L. “Cooperation between two IEEE International Multiple Sensors, In this paper, an absolute localization paradigm based on the cooperation
Delahoche, et al. omnidirectional perception Conference on Intelligent Robotics of an omnidirectional vision system composed of a conical mirror and a
systems for mobile robot Robots and Systems 2: CCD camera and a low cost panoramic range finder system is reported.
localization.” 1499-1504 (2000). We present an absolute localization method that uses three matching
criteria fused by the combination rules of the Dempster-Shafer theory.

Graham, J. H. “Sensory-Based International Journal of

Robotics, Sensors, This paper presents a multilevel system for contributing to robot safety
Safeguarding of Robotic Robotics & Automation Decision-Making by the use of sensory information in partially defined environments,
Systems.” 9(4): 141-148, (1994). including provision for sensory preprocessing, sensory fusion, and high-
level decision making. Sensory fusion is achieved by using Dempster's
rule of combination on a set of belief fuctions generated from the input
sensory data.
Hughes, K. and “Ultrasonic robot Proceedings of SPIE The Robotics, Sensors In this paper we present a method for ultrasonic robot localization
R. R. Murphy localization using International Society for without a priori world models utilizing the ideas of distinctive places and
Dempster-Shafer theory.” Optical Engineering: 2-11, open space attraction. This method was incorporated into a move-to-
(1992). station behavior, which was demonstrated on the Georgia Tech mobile
robot. The key aspect of our approach was to use Dempster-Shafer
theory to overcome the problem of the uncertainty in the range
measurements returned by the sensors.

Luo, Z. and D. Li “Multi-source information IEEE International Robotics, Artificial In the paper, we develop a new multisource information fusion scheme
integration in intelligent Conference on Multisensor Vision, Object using the plausibility measure. The method avoids using Dempster's rule
systems using the Fusion and Integration for Recognition, of combination, so as to overcome the intractability of Dempster-Shafer
plausibility measure.” Intelligent Systems: 403- Sensors computations, allowing the theory to be feasible in many more
409, (1994). applications. A simple robotic vision system with object recognition data
from multisensor is presented to highlight benefits of the new method.

Murphy, R. R. “Dempster-Shafer theory IEEE Transactions on Multiple Sensors, This article discusses Dempster-Shafer (DS) theory in terms of its utility
for sensor fusion in Robotics and Automation Robotics for sensor fusion for autonomous mobile robots, It exploits two little
autonomous mobile 14(2): 197-206 (1998). used components of DS theory: the weight of conflict metric and the
robots.” enlargement of the frame of discernment.

Appendix A-31
 ROBOTICS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Murphy, R. R. “Estimating time available Muliple Sensors, In this paper, we consider the impact of time for teleoperation
and E. Rogers for sensor fusion exception Robotics applications where a remote robot attempts to autonomously maintain
handling.” sensing in the presence of failures yet has the option to contact the local
for further assistance. Time limits are determined by using evidential
reasoning with a novel generalization of Dempster-Shafer theory.

Puente, E. A., L. “Analysis of data fusion IECON Proceedings 2: Robotics, The authors focus on the use of the occupancy grid representation to
Moreno, et al. methods in certainty grids 1133-1137, (1991). Monitoring maintain and combine the information acquired from sensors about the
application to collision environment. This information is subsequently used to monitor the robot
danger monitoring.” collision danger risk and take into account that risk in starting the
appropriate maneuver. The occupancy grid representation uses a
multidimensional tessellation of space into cells, where each cell stores
some information about its state. Two main approaches have been used
to model the occupancy of a cell: probabilistic estimation and the
Dempster-Shafer theory of evidence. Probabilistic estimation and some
combination rules based on the Dempster-Shafer theory of evidence are
analyzed and their possibilities compared.

Ribo, M. and A. “A comparison of three Robotics and Autonomous Robotics, Sensors In this paper, we describe and compare three different uncertainty calculi
Pinz uncertainty calculi for Systems 35(3-4): 201-209, techniques to build occupancy grids of an unknown environment using
building sonar-based (2001). sensory information provided by a ring of ultrasonic range-finders. These
occupancy grids.” techniques are based on Bayesian theory, Dempster-Shafer theory of
evidence, and fuzzy set theory.

Appendix A-32
 ROBOTICS (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Utete, S. W., B. “Voting as validation in International Journal of Robotics, Target This paper investigates the use of voting as a conflict-resolution
Barshan, et al. robot programming.” Robotics Research 18(4): Identification, technique for data analysis in robot programming. Dispersed sensors take
401-413, (1999). Sensors, Decision- decisions on target type, which must then be fused to give the single
Making group classification of the presence or absence and type of a target
Dempster-Shafer evidential reasoning is used to assign a level of belief to
each sensor decision. The decisions are then fused by two means. Using
Dempster's rule of combination, conflicts are resolved through a group
measure expressing dissonance in the sensor views.

Wu, Y., J. Huang, “Mobile robot obstacle Zidonghua Xuebao/Acta Robotics, Modeling 2D environment and road detection for mobile robot by fusing
et al. detection and environment Automatica Sinica 23: 641- Environment color and range image information are discussed. The environment
modeling with sensor 648, (1997). Detection, model is constructed by using multi-resolution 2D grid representation,
fusion.” which is proved to be a better solution to the tradeoff between accuracy
and computation speed. The fusion algorithm is designed based on a
generalized Dempster-Shafer theory of evidence (DSTE), which is
efficient in dealing with dependent information.

SIGNAL DETECTION AND PROCESSING

AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Boston, J. R. “Signal Detection Models Uncertainty Modelling and Classification, This chapter develops models for signal detection in noisy waveforms,
Incorporating Uncertainty: Analysis: Theory and Signal Detection, based on Dempster-Shafer theory and on fuzzy logic, that classify
Sensitivity to Parameter Applications. B. M. Ayyub Sensitivity Analysis waveforms as signal-present, signal-absent, or uncertain. The
Estimates.” and M. M. Gupta. New performances of the models were evaluated using simulated sensory
York, Elsevier. 17: 459- evoked potential data and compared to a Bayesian maximum likelihood
476(1994). detector. The effects of errors in estimates of the statistical parameters of
the wave forms are considered.

Appendix A-33
 SIGNAL DETECTION AND PROCESSING (continued)
AUTHOR(S) TITLE REFERENCE GENERAL APPLICATION
SUBJECT
HEADINGS
Boston, J. R. “A signal detection system IEEE Transactions on Classification, This paper describes a signal detection algorithm based on Dempster-
based on Dempster-Shafer Systems Man and Signal Detection Shafer theory: The detector combines evidence provided by multiple
theory and comparison to Cybernetics Part C- waveform features and explicitly considers uncertainty in the detection
fuzzy detection.” Applications and Reviews decision, The detector classifies waveforms as including a signal, not
30(1): 45-51, (2000). including a signal, or being uncertain, in which case no conclusion
regarding presence or absence of a signal is drawn.

Chao, J. J. and C. “An Efficient Direct- IEEE Transactions on

C. Lee Sequence Signal Detector Communications 38(6): 868-
Based on Dempster-Shafer 874, (1990).
Theory.”
Chao, J. J., C. M. “A moving target detector Target Detection, Moving target detector (MTD) related multiple-hypothesis testing is
Cheng, et al. based on information Radar, Signals considered, and the Dempster-Shafer theory is applied to this problem.
fusion.” Feature parameters are extracted from radar signals, and the value of
each feature parameter is interpreted in terms of Dempster-Shafer's belief
or disbelief for the associated hypotheses. Using Dempster's combining
rule, a generalized likelihood ratio test is derived.

Hughes, R. C. “Acoustic Signal Pattern Recognition 18: 475- Identification, An expert system approach to identifying the sources of underwater
and J. N. Interpretation Reasoning 483, (1984). Expert Systems acoustic signals is described. In order to deal with non-specific and
Maksym with Non-Specific and uncertain evidence in the presence of an unknown number of signal
Uncertain Information.” sources, we develop an inference network approach which is based on
the Dempster-Shafer theory of evidence.

Jang, L.-W. and “Information fusion Target Tracking, We employ the technique of uncertain information processing to solve
J.-J. Chao algorithm for data Radar, Sonar problems of multitarget tracking. We consider the data association
association in multitarget problem as a fuzzy partition. Dempster-Shafer theory is used to evaluate
tracking.” the plausibilities of the association events. Using the plausibilities, a
fuzzy partition is performed.

Appendix A-34
External Distribution

Bilal Ayyub (2) Thomas A. Cruse

Department of Civil Engineering 398 Shadow Place
University of Maryland Pagosa Springs, CO 81147-7610
College Park, MD 20742
Department of Energy (3)
Ivo Babuska Attn: William Reed, NA-114
Texas Institute for Computational Jamileh Soudah, NA-114
and Applied Mathematics B. Pate, NA-114
Mail Code C0200 1000 Independence Ave., SW
University of Texas at Austin Washington, DC 20585
Austin, TX 78712-1085
U. M. Diwekar (2)
Osman Balci Center for Energy and
Department of Computer Science Environmental Studies
Virginia Tech Carnegie Mellon University
Blacksburg, VA 24061 Pittsburgh, PA 15213-3890

Steven Batill (2) Robert G. Easterling

Dept. of Aerospace & Mechanical Engr. 7800 Northridge NE
University of Notre Dame Albuquerque, NM 87109
Notre Dame, IN 46556
Doug Elias
James Berger Parker Center for Investment Research
Inst. of Statistics and Decision Science Cornell University
Duke University 302 Sage Hall
Box 90251 Ithaca, NY 14853-6201
Durham, NC 27708-0251
Isaac Elishakoff
Chun-Hung Chen (2) Dept. of Mechanical Engineering
Department of Systems Engineering & Florida Atlantic University
Operations Research 777 Glades Road
George Mason University Boca Raton, FL 33431-0991
4400 University Drive, MS 4A6
Fairfax, VA 22030 Ashley Emery
Dept. of Mechanical Engineering
Raymond Cosner (2) Box 352600
Boeing-Phantom Works University of Washingtion
MC S106-7126 Seattle, WA 98195-2600
P. O. Box 516
St. Louis, MO 63166-0516

87
C. Robert Emerson Raphael Haftka (2)
Department of Systems Science and Dept. of Aerospace and Mechanical
Industrial Engineering Engineering and Engr. Science
Binghamton University P. O. Box 116250
P.O. Box 6000 University of Florida
Binghamton, NY 13902-6000 Gainesville, FL 32611-6250

Scott Ferson (5) Achintya Haldar (2)

Applied Biomathematics Dept. of Civil Engineering
100 North Country Road & Engineering Mechanics
Setauket, New York 11733-1345 University of Arizona
Tucson, AZ 85721
Mike Giltrud
Defense Threat Reduction Agency Tim Hasselman
DTRA/CPWS ACTA
6801 Telegraph Road 2790 Skypark Dr., Suite 310
Alexandria, VA 22310-3398 Torrance, CA 90505-5345

James Glimm (2) George Hazelrigg

Dept. of Applied Math & Statistics Division of Design, Manufacturing
P138A & Innovation
State University of New York Room 508N
Stony Brook, NY 11794-3600 4201 Wilson Blvd.
Arlington, VA 22230
James Gran
SRI International David Higdon
Poulter Laboratory AH253 Inst. of Statistics and Decision Science
333 Ravenswood Avenue Duke University
Menlo Park, CA 94025 Box 90251
Durham, NC 27708-0251
Bernard Grossman (2)
Dept. of Aerospace & Richard Hills (2)
Ocean Engineering College of Engineering, MSC 3449
Mail Stop 0203 New Mexico State University
215 Randolph Hall P. O. Box 30001
Blacksburg, VA 24061 Las Cruces, NM 88003

Sami Habchi F. Owen Hoffman (2)

CFD Research Corp. SENES
Cummings Research Park 102 Donner Drive
215 Wynn Drive Oak Ridge, TN 37830
Huntsville, AL 35805

88
Luc Huyse Robert Lust
Southwest Research Institute General Motors, R&D and Planning
P. O. Drawer 28510 MC 480-106-256
San Antonio, TX 78284-0510 30500 Mound Road
Warren, MI 48090-9055
Leo Kadanoff (2)
Research Institutes Building Sankaran Mahadevan (2)
University of Chicago Dept. of Civil &
5640 South Ellis Ave. Environmental Engineering
Chicago, IL 60637 Vanderbilt University
Box 6077, Station B
George Karniadakis (2) Nashville, TN 37235
Division of Applied Mathematics
Brown University Hans Mair
192 George St., Box F Institute for Defense Analysis
Providence, RI 02912 Operational Evaluation Division
1801 North Beauregard Street
Alan Karr Alexandria, VA 22311-1772
Inst. of Statistics and Decision Science
Duke University Gregory McRae (2)
Box 90251 Dept. of Chemical Engineering
Durham, NC 27708-0251 Massachusetts Institute of Technology
Cambridge, MA 02139
George Klir (2)
Department of Systems Science and Michael Mendenhall (2)
Industrial Engineering Nielsen Engineering & Research, Inc.
Binghamton University 510 Clyde Ave.
P.O. Box 6000 Mountain View, CA 94043
Binghamton, NY 13902-6000
Sue Minkoff (2)
Sarah Lam Dept. of Mathematics and Statistics
Department of Systems Science and University of Maryland
Industrial Engineering 1000 Hilltop Circle
Binghamton University Baltimore, MD 21250
P.O. Box 6000
Binghamton, NY 13902-6000 Max Morris (2)
Department of Statistics
Harold Lewis III Iowa State University
Department of Systems Science and 304A Snedecor-Hall
Industrial Engineering Ames, IW 50011-1210
Binghamton University
P.O. Box 6000 NASA/Ames Research Center (2)
Binghamton, NY 13902-6000 Attn: Unmeel Mehta, MS T27 B-1
David Thompson, MS 269-1
Moffett Field, CA 94035-1000
89
NASA/Glen Research Center (2)
Attn: John Slater, MS 86-7 Pradeep Raj (2)
Chris Steffen, MS 5-11 Computational Fluid Dynamics
21000 Brookpark Road Lockheed Martin Aeronautical Sys.
Cleveland, OH 44135 86 South Cobb Drive
Marietta, GA 30063-0685
NASA/Langley Research Center (7)
Attn: Dick DeLoach, MS 236 John Renaud (2)
Michael Hemsch, MS 280 Dept. of Aerospace & Mechanical Engr.
Tianshu Liu, MS 238 University of Notre Dame
Jim Luckring, MS 280 Notre Dame, IN 46556
Joe Morrison, MS 128
Ahmed Noor, MS 369 Tim Ross (2)
Sharon Padula, MS 159 Dept. of Civil Engineering
Hampton, VA 23681-0001 University of New Mexico
Albuquerque, NM 87131
Robert Nelson
Dept. of Aerospace & Mechanical Engr. J. Sacks
University of Notre Dame Inst. of Statistics and Decision Science
Notre Dame, IN 46556 Duke University
Box 90251
Dick Neumann Durham, NC 27708-0251
8311 SE Millihanna Rd.
Olalla, WA 98359 Sunil Saigal (2)
Carnegie Mellon University
Efstratios Nikolaidis (2) Department of Civil and
MIME Dept. Environmental Engineering
4035 Nitschke Hall Pittsburgh, PA 15213
University of Toledo
Toledo, OH 43606-3390 Len Schwer
Schwer Engineering & Consulting
Dale Pace 6122 Aaron Court
Applied Physics Laboratory Windsor, CA 95492
Johns Hopkins University
111000 Johns Hopkins Road Paul Senseny
Laurel, MD 20723-6099 Factory Mutual Research Corporation
1151 Boston-Providence Turnpike
Alex Pang P.O. Box 9102
Computer Science Department Norwood, MA 02062
University of California
Santa Cruz, CA 95064

Allan Pifko
2 George Court
Melville, NY 11747
90
Kari Sentz (50) Bill Spencer (2)
Department of Systems Science and Dept. of Civil Engineering
Industrial Engineering and Geological Sciences
Binghamton University University of Notre Dame
P.O. Box 6000 Notre Dame, IN 46556-0767
Binghamton, NY 13902-6000
D. E. Stevenson (2)
E. Sevin Computer Science Department
Logicon RDA, Inc. Clemson University
1782 Kenton Circle 442 Edwards Hall, Box 341906
Lyndhurst, OH 44124 Clemson, SC 29631-1906

Mark Shephard (2) Ben Thacker

Rensselaer Polytechnic Institute Southwest Research Institute
Scientific Computation Research Center P. O. Drawer 28510
Troy, NY 12180-3950 San Antonio, TX 78284-0510

Tom I-P. Shih Fulvio Tonon (2)

Dept. of Mechanical Engineering Geology and Geophysics Dept.
2452 Engineering Building East Room 719
East Lansing, MI 48824-1226 University of Utah
135 South 1460
T. P. Shivananda Salt Lake City, UT 84112
Bldg. SB2/Rm. 1011
TRW/Ballistic Missiles Division Robert W. Walters (2)
P. O. Box 1310 Aerospace and Ocean Engineering
San Bernardino, CA 92402-1310 Virginia Tech
215 Randolph Hall, MS 203
Don Simons Blacksburg, VA 24061-0203
Logicon
222 W. Sixth St. Justin Y-T Wu
P.O. Box 471 Applied Research Associates
San Pedro, CA 90733-0471 Probabilistic Engineering
811 Spring Forest Rd.
Ashok Singhal Raleigh, NC 27609
CFD Research Corp.
Cummings Research Park Ren-Jye Yang
215 Wynn Drive Ford Research Laboratory
Huntsville, AL 35805 MD2115-SRL
P.O.Box 2053
Dearborn, MI 4812

91
Simone Youngblood (2)
DOD/DMSO
Technical Director for VV&A
1901 N. Beauregard St., Suite 504
Alexandria, VA 22311

92
Foreign Distribution

Yakov Ben-Haim (2)

Department of Mechanical Engineering K. Papoulia
Technion-Israel Institute of Technology Inst. Eng. Seismology & Earthquake
Haifa 32000 Engineering
ISRAEL P.O. Box 53, Finikas GR-55105
Thessaloniki
Gert de Cooman (2) GREECE
Universiteit Gent
Onderzoeksgroep, SYSTeMS Max Ruppert
Technologiepark - Zwijnaarde 9 UniBw Munich - BauV 2.2
9052 Zwijnaarde Inst. Engng.Mech. & Struct.Mech.
BELGIUM D - 85577 Neuibiberg
GERMANY
Luis Eca (2)
Instituto Superior Tecnico Lev Utkin
Department of Mechanical Engineering Institute of Statistics
Av. Rovisco Pais Munich University
1096 Lisboa CODEX Ludwigstr. 33
PORTUGAL 80539, Munich
GERMANY
Charles Hirsch (2)
Department of Fluid Mechanics Malcolm Wallace
Vrije Universiteit Brussel National Engineering Laboratory
Pleinlaan, 2 East Kilbride
B-1050 Brussels Glasgow
BELGIUM G75 0QU
UNITED KINGDOM
Igor Kozine (2)
Systems Analysis Department Peter Walley
Riso National Laboratory 6 Jewel Close
P. O. Box 49 Port Douglas
DK-4000 Roskilde Queensland 4871
DENMARK AUSTRALIA

93
Department of Energy Laboratories

Los Alamos National Laboratory (13)

Mail Station 5000
P.O. Box 1663
Los Alamos, NM 87545
Attn: Jane M. Booker, MS P946
Terrence Bott, MS K557
Scott Doebling, MS P946
S. Eisenhawer, MS K557
James Hyman, MS B284
Cliff Joslyn, MS B265
S. Keller-McNulty, MS F600
Deborah Leishman, MS F600
Mike McKay, MS F600
Laura McNamara, MS F600
Karen I. Pao, MS B256
Alyson G. Wilson, MS F600
Gregory Wilson, MS F600

Lawrence Livermore National Laboratory (3)

7000 East Ave.
P.O. Box 808
Livermore, CA 94550
Attn: Richard Klein, MS L-023
Roger Logan, MS L-125
Cynthia Nitta, MS L-096

Argonne National Laboratory

Attn: Paul Hovland
MCS Division
Bldg. 221, Rm. C-236
9700 S. Cass Ave.
Argonne, IL 60439

94
Sandia Internal Distribution

1 MS 0427 2104 F. F. Dean 1 MS 0847 9120 H. S. Morgan

1 MS 0482 2109 S. E. Lott 1 MS 0555 9122 M. S. Garrett
1 MS 0759 5845 I. V. Waddoups 1 MS 0893 9123 R. M. Brannon
1 MS 0751 6117 L. S. Costin 1 MS 0847 9124 J. M. Redmond
1 MS 0708 6214 P. S. Veers 1 MS 0847 9124 K. F. Alvin
1 MS 0490 6252 J. A. Cooper 1 MS 0553 9124 T. G. Carne
1 MS 0747 6410 A. L. Camp 1 MS 0847 9124 R. V. Field
1 MS 0747 6410 G. D. Wyss 1 MS 0847 9124 S. F. Wojtkiewicz
1 MS 0748 6413 D. G. Robinson 1 MS 0557 9125 T. J. Baca
1 MS 0748 6413 R. D. Waters 1 MS 0847 9126 R. A. May
1 MS 0451 6502 T. D. Woodall 1 MS 0824 9130 J. L. Moya
1 MS 1137 6534 S. M. DeLand 1 MS 1135 9132 L. A. Gritzo
1 MS 1137 6536 G. K. Froehlich 1 MS 0828 9133 M. Pilch
1 MS 1137 6536 A. L. Hodges 1 MS 0828 9133 B. F. Blackwell
1 MS 0716 6804 P. G. Kaplan 1 MS 0828 9133 K. J. Dowding
1 MS 1395 6820 D. K. Belasich 40 MS 0828 9133 W. L. Oberkampf
1 MS 1395 6820 M. J. Chavez 1 MS 0557 9133 T. L. Paez
1 MS 1395 6820 J. G. Miller 1 MS 0828 9133 V. J. Romero
1 MS 1395 6821 M. K. Knowles 1 MS 0828 9133 M. P. Sherman
1 MS 1395 6821 J. W. Garner 1 MS 0557 9133 A. Urbina
1 MS 1395 6821 E. R. Giambalvo 1 MS 0847 9133 W. R. Witkowski
1 MS 1395 6821 T. Hadgu 1 MS 1135 9134 S. Heffelfinger
1 MS 1395 6821 S. C. James 1 MS 0835 9140 J. M. McGlaun
1 MS 1395 6821 J. S. Stein 1 MS 0835 9141 S. N. Kempka
1 MS 0779 6840 M. G. Marietta 1 MS 0835 9142 J. S. Peery
1 MS 0779 6840 P. Vaughn 1 MS 0827 9143 J. D. Zepper
2 MS 0779 6849 J. C. Helton 1 MS 1110 9214 D. E. Womble
1 MS 0779 6849 L. C. Sanchez 1 MS 0847 9211 M. S. Eldred
1 MS 0778 6851 G. E. Barr 1 MS 0847 9211 A. A. Giunta
1 MS 0778 6851 R. J. MacKinnon 1 MS 0847 9211 J. R. Red-Horse
1 MS 0778 6851 P. N. Swift 1 MS 0819 9211 T. G. Trucano
1 MS 0779 6852 B. W. Arnold 1 MS 1110 9214 R. B. Lehoucq
1 MS 0779 6852 R. P. Rechard 1 MS 0310 9220 R. W. Leland
1 MS 9404 8725 J. R. Garcia 1 MS 0318 9230 P. Yarrington
1 MS 9405 8726 R. E. Jones 1 MS 0819 9231 E. A. Boucheron
1 MS 9042 8727 J. J. Dike 1 MS 0316 9233 S. S. Dosanjh
1 MS 9012 8920 P. E. Nielan 1 MS 0316 9235 J. B. Aidun
1 MS 9003 8950 C. M. Hartwig 1 MS 0660 9519 D. S. Eaton
1 MS 0841 9100 T. C. Bickel 1 MS 0660 9519 M. A. Ellis
1 MS 0825 9115 W. H. Rutledge 1 MS 0139 9900 M. O. Vahle
1 MS 0836 9116 E. S. Hertel 1 MS 0139 9904 R. K. Thomas
1 MS 0836 9116 D. Dobranich 1 MS 0428 12300 D. D. Carlson
1 MS 0836 9117 R. O. Griffith 1 MS 0428 12301 V. J. Johnson
95
1 MS 0421 12323 J. M. Sjulin 1 MS 1176 15312 R. M. Cranwell
1 MS 0829 12323 B. M. Rutherford 1 MS 1176 15312 D. J. Anderson
1 MS 0829 12323 F. W. Spencer 1 MS 1176 15312 J. E. Campbell
1 MS 0638 12326 M. A. Blackledge 1 MS 1176 15312 L. P. Swiler
1 MS 0638 12326 D. E. Peercy 1 MS 1179 15340 J. R. Lee
1 MS 0638 12326 D. L. Knirk 1 MS 1179 15341 L. Lorence
1 MS 0492 12332 D. R. Olson 1 MS 9018 8945-1 Central Technical
1 MS 0405 12333 T. R. Jones Files
1 MS 0405 12333 S. E. Camp 2 MS 0899 9616 Technical Library
1 MS 0434 12334 R. J. Breeding 1 MS 0612 9612 Review & Approval
1 MS 0829 12335 K. V. Diegert Desk For DOE/OSTI
1 MS 1170 15310 R. D. Skocypec

96