0% found this document useful (0 votes)
105 views14 pages

Assignment2 (Section B)

This document provides the calculations for backpropagation learning over two iterations on a 3-2-2 neural network with two input patterns. In the first iteration, the network is run forward on Pattern 1 to calculate outputs, errors, and weight changes. Weights are then updated without momentum. In the second iteration, the network is run on Pattern 2 to again calculate outputs, errors, and weight changes, with weights now updated using the learning rate and momentum from the first iteration.

Uploaded by

Nur Syuhada
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
105 views14 pages

Assignment2 (Section B)

This document provides the calculations for backpropagation learning over two iterations on a 3-2-2 neural network with two input patterns. In the first iteration, the network is run forward on Pattern 1 to calculate outputs, errors, and weight changes. Weights are then updated without momentum. In the second iteration, the network is run on Pattern 2 to again calculate outputs, errors, and weight changes, with weights now updated using the learning rate and momentum from the first iteration.

Uploaded by

Nur Syuhada
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

FACULTY OF INFORMATION & COMMUNICATION

TECHNOLOGY
UNIVERSITI TEKNIKAL MALAYSIA MELAKA

BITI 3133 NEURAL NETWORK


ASSIGNMENT 2 SECTION B

STUDENT’S : NUR SYUHADA BINTI AZHAR


NAME
MATRIC NO. : B032010378
SECTION : S1G2

LECTURER : PROFESSOR TS. DR. BURHANUDDIN BIN MOHD


ABOOBAIDER
QUESTION:
Assume a standard 3-2-2 three-layered network with η = 1.1 and α= 0.6. Further assume that
intitial values for all biases are 0.005, the values of input to hidden weights are -0.27 and hidden
to output weights are 0.31. Perform calculations using backpropagation learning for two
iterations on the pattern sets as given:

k 𝒙𝒌𝟏 𝒙𝒌𝟐 𝒙𝒌𝟑 𝒅𝒌𝟏 𝒅𝒌𝟐


1 0.5 0.1 0.3 1 0
2 0.4 0.6 0.4 0 1

Pattern 1: Given η = 1.1 and momentum, α= 0.6.

0.005 0.005
0.5

0.31

0.1

0.31

0.3
0.005 0.005

Forward pass on pattern 1:

z11 = [x11 × m111 ] + [x12 × m121 ] + [x13 × m131 ] + m101

= [0.5 x (−0.27)] + [0.1 x (−0.27)] + [0.3 x (−0.27)] + 0.005


= (−0.238)

z21 = [x11 × m112 ] + [x12 × m122 ] + [x13 × m132 ] + m102

= [0.5 x (−0.27)] + [0.1 x (−0.27)] + [0.3 x (−0.27)] + 0.005


= (−0.238)
1
S(z11 ) = [ 1 ]
1+e−z1

1
= [1+e0.238]

= 0.4408

1
S(z21 ) = [ 1 ]
1+e−z2

1
= [1+e0.238]

= 0.4408

y11 = [S(z21 ) × n111 ] + [S(z21 ) × n121 ] + n101

= [0.4408 × 0.31] + [0.4408 × 0.31] + 0.005

= 0.2783

y21 = [S(z21 ) × n112 ] + [S(z21 ) × n122 ] + n102

= [0.4408 × 0.31] + [0.4408 × 0.31] + 0.005

= 0.2783

1
S(y11 ) = [ 1 ]
1+e−y1

1
= [1+e−0.2783 ]

= 0.5691

1
S(y21 ) = [ 1 ]
1+e−y2

1
= [1+e−0.2783 ]

= 0.5691

e11 = d11 − S(y11 )

= 1 – 0.5691

= 0.4309
e12 = d12 − S(y21 )

= 0 – 0.5691

= (-0.5691)

δ11 = e11 × S(z11 )[1 − S(z11 )]

= 0.4309 × 0.4408(1 − 0.4408)

= 0.1062

δ12 = e12 × S(z21 )[1 − S(z21 )]

= −0.5691 × 0.4408(1 − 0.4408)

= (−0.1403)

δH11 = [(δ11 × n111 ) + (δ12 × n112 )] × [S(z11 )[1 − S(z11 )]]

= [(0.1062 × 0.31) + (−0.1403 × 0.31)] × [0.4408[1 − 0.4408]]

= (−0.0026)

δH21 = [(δ11 × n121 ) + (δ12 × n122 )] × [S(z21 )[1 − S(z21 )]]

= [(0.1062 × 0.31) + (−0.1403 × 0.31)] × [0.4408[1 − 0.4408]]

= (−0.0026)

Weight changes are affected with η = 1.1;


∆m101 = η × δH11 × 1

= 1.1 × (−0.0026) × 1

= (−0.0029)

∆m111 = η × δH11 × S(x11 )

= 1.1 × (−0.0026) × 0.5

= (−0.0014)
∆m121 = η × δH11 × S(x12 )

= 1.1 × (−0.0026) × 0.1

= (−0.0003)
∆m131 = η × δH11 × S(x13 )

= 1.1 × (−0.0026) × 0.3

= (−0.0009)

∆m102 = η × δH21 × 1

= 1.1 × (−0.0028) × 1

= (−0.0031)

∆m112 = η × δH21 × S(x11 )

= 1.1 × (−0.0026) × 0.5

= (−0.0014)

∆m122 = η × δH21 × S(x12 )

= 1.1 × (−0.0026) × 0.1

= (−0.0003)

∆m132 = η × δH21 × S(x12 )

= 1.1 × (−0.0026) × 0.3

= (−0.0009)

∆n101 = η × δ11 × 1

= 1.1 × 0.1062 × 1

= 0.1168

∆n111 = η × δ11 × S(z11 )

= 1.1 × 0.1062 × 0.4408

= 0.0515
∆n121 = η × δ11 × S(z21 )

= 1.1 × 0.1062 × 0.4408

= 0.0515

∆n102 = η × δ12 × 1

= 1.1 × (−0.1403) × 1

= (−0.1543)

∆n112 = η × δ12 × S(z11 )

= 1.1 × (−0.1403) × 0.4408

= (−0.0680)

∆n122 = η × δ12 × S(z21 )

= 1.1 × (−0.1403) × 0.4408

= (−0.0680)

Weights are modified;

n201 = 𝑛101 + ∆𝑛101

= 0.005 + 0.1168

= 0.1218

2 1 1
n11 = 𝑛11 + ∆𝑛11

= 0.31 + 0.0515

= 0.3615

n221 = 𝑛121 + ∆𝑛121

= 0.31 + 0.0515

= 0.0565
n202 = 𝑛102 + ∆𝑛102

= 0.005 + (−0.1543)

= (−0.1493)

2 1 1
n12 = 𝑛12 + ∆𝑛12

= 0.31 + (−0.0680)

= 0.242

n222 = 𝑛122 + ∆𝑛122

= 0.31 + (−0.0680)

= 0.242

2
𝑚01 = 𝑚101 + ∆𝑚101

= 0.005 + (−0.0029)

= (0.0021)

2 1 1
𝑚11 = 𝑚11 + ∆𝑚11

= (−0.27) + (−0.0014)

= (−0.2714)

2
𝑚21 = 𝑚121 + ∆𝑚121

= (−0.27) + (−0.0003)

= (−0.2703)

2
𝑚31 = 𝑚131 + ∆𝑚131

= (−0.27) + (−0.0009)

= (−0.2709)

2
𝑚02 = 𝑚102 + ∆𝑚102

= 0.005 + (−0.0029)

= (0.0021)
2 1 1
𝑚12 = 𝑚12 + ∆𝑚12

= (−0.27) + (−0.0014)

= (−0.2714)

2
𝑚22 = 𝑚122 + ∆𝑚122

= (−0.27) + (−0.0003)

= (−0.2703)

2
𝑚32 = 𝑚132 + ∆𝑚132

= (−0.27) + (−0.0009)

= (−0.2709)

Notice that since this is the first iteration there is no effect of the momentum term since the
weight change on the previous iteration was zero.

Pattern 2:

0.0021 0.1218
0.4

0.3615

0.6

0.242

0.4
0.0021 -0.1493

Forward pass on pattern 2 :

z12 = [x12 × m11


2 ]
+ [x22 × m221 ] + [x32 × m231 ] + m201

= [0.4 × (−0.2714)] + [0.6 × (−0.2703)] + [0.4 × (−0.2709)] + 0.0021

= (-0.377)
z22 = [x12 × m12
2 ]
+ [x22 × m222 ] + [x32 × m232 ] + m202

= [0.4 × (−0.2714)] + [0.6 × (−0.2703)] + [0.4 × (−0.2709)] + 0.0021

= (-0.377)

1
S(z12 ) = [ 2 ]
1+e−z1

1
= [1+e0.377]

= 0.4069

1
S(z22 ) = [ 2 ]
1+e−z1

1
= [1+e0.377]

= 0.4069

y12 = [S(z12 ) × n11


2 ]
+ [S(z12 ) × n221 ] + n201

= [0.4069 × 0.3615] + [0.4069 × 0.3615] + 0.1218

= 0.4160

y22 = [S(z22 ) × n12


2 ]
+ [S(z22 ) × n222 ] + n202

= [0.4069 × 0.242] + [0.4069 × 0.242] + (−0.1493)

= 0.0476

1
S(y12 ) = [ 2 ]
1+e−y1

1
= [1+e−0.4160 ]

= 0.6025

1
S(y22 ) = [ 2 ]
1+e−y2

1
= [1+e−0.0476 ]

= 0.5119
e12 = d12 − S(y12 )

= 0 – 0.6025

= (-0.6025)

e22 = d22 − S(y22 )

= 1 – 0.5119

= 0.4881

δ12 = e12 × S(z12 )[1 − S(z12 )]

= (−0.6025 × 0.4069(1 − 0.4069)

= (−0.1454)

δ22 = e22 × S(z22 )[1 − S(z22 )]

= 0.4881 × 0.4069(1 − 0.4069)

= 0.1178

δH12 = [(δ12 × n11


2 )
+ (δ22 × n12
2 )]
× [S(z12 )[1 − S(z12 )]]

= [(−0.1454 × 0.3615) + (0.1178 × 0.242)] × [0.4069[1 − 0.4069]]

= (−0.0457)

δH22 = [(δ12 × n221 ) + (δ22 × n222 )] × [S(z22 )[1 − S(z22 )]]

= [(−0.1454 × 0.3615) + (0.1178 × 0.242)] × [0.4069[1 − 0.4069]]

= (−0.0457)

Weight changes are affected with η = 1.1;


∆m201 = η × δH12 × 1

= 1.1 × (−0.0457) × 1

= (−0.0503)
2
∆m11 = η × δH12 × S(x12 )

= 1.1 × (−0.0457) × 0.4

= (−0.0201)

∆m221 = η × δH12 × S(x22 )

= 1.1 × (−0.0457) × 0.6

= (−0.0302)

∆m231 = η × δH12 × S(x32 )

= 1.1 × (−0.0457) × 0.4

= (−0.0201)

∆m202 = η × δH22 × 1

= 1.1 × (−0.0457) × 1

= (−0.0503)

2
∆m12 = η × δH22 × S(x12 )

= 1.1 × (−0.0457) × 0.4

= (−0.0201)

∆m222 = η × δH22 × S(x22 )

= 1.1 × (−0.0457) × 0.6

= (−0.0302)

∆m232 = η × δH22 × S(x22 )

= 1.1 × (−0.0457) × 0.4

= (−0.0201)

∆n201 = η × δ12 × 1

= 1.1 × (−0.1454) × 1

= (−0.1599)
2
∆n11 = η × δ12 × S(z12 )

= 1.1 × (−0.1454) × 0.4069

= (−0.0651)

∆n221 = η × δ12 × S(z22 )

= 1.1 × (−0.1454) × 0.4069

= (−0.0651)

∆n202 = η × δ22 × 1

= 1.1 × 0.1178 × 1

= 0.1296

2
∆n12 = η × δ22 × S(z12 )

= 1.1 × 0.1178 × 0.4069

= 0.0527

∆n222 = η × δ22 × S(z22 )

= 1.1 × 0.1178 × 0.4069

= 0.0527

Weights are modified;

Since there were weights changes in the previous iteration, the momentum term comes into
the update procedure.

n301 = n201 + ∆n201 + ∝ × ∆n101

= 0.1218 + (−0.1599) + 0.6 × 0.1168

= 0.0320

3 2 2
n11 = n11 + ∆n11 + ∝ × ∆n111

= 0.3615 + (−0.0651) + 0.6 × 0.0515

= 0.3273
n321 = n221 + ∆n221 + ∝ × ∆n121

= 0.3615 + (−0.0651) + 0.6 × 0.0515

= 0.3273

n302 = n202 + ∆n202 + ∝ × ∆n102

= (−0.1493) + 0.1296 + 0.6 × (−0.1543)

= (−0.1123)

3 2 2
n12 = n12 + ∆n12 + ∝ × ∆n112

= 0.242 + 0.0527 + 0.6 × (−0.0680)

= 0.2539

n322 = n222 + ∆n222 + ∝ × ∆n122

= 0.242 + 0.0527 + 0.6 × (−0.0680)

= 0.2539

m301 = m201 + ∆m201 + ∝ × ∆m101

= 0.0021 + (−0.0503) + 0.6 + (−0.0029)

= (−0.0499)

3 2 2
m11 = m11 + ∆m11 + ∝ × ∆m111

= (−0.2714) + (−0.0201) + 0.6 × (−0.0014)

= (−0.2923)

m321 = m221 + ∆m221 + ∝ × ∆m121

= (−0.2703) + (−0.0302) + 0.6 × (−0.0003)

= (−0.3007)

m331 = m231 + ∆m231 + ∝ × ∆m131

= (−0.2709) + (−0.0201) + 0.6 × (−0.0009)

= (−0.2915)
m302 = m202 + ∆m202 + ∝ × ∆m102

= 0.0021 + (−0.0503) + 0.6 + (−0.0029)

= (−0.0499)

3 2 2
m12 = m12 + ∆m12 + ∝ × ∆m112

= (−0.2714) + (−0.0201) + 0.6 × (−0.0014)

= (−0.2923)

m322 = m222 + ∆m222 + ∝ × ∆m122

= (−0.2703) + (−0.0302) + 0.6 × (−0.0003)

= (−0.3007)

m332 = m232 + ∆m232 + ∝ × ∆m132

= (−0.2709) + (−0.0201) + 0.6 × (−0.0009)

= (−0.2915)

Pattern 3:

-0.0499 0.0320

0.3273

0.2539

-0 .0499 -0.1123

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy