Scribd
Upload
8 views18 pages

1 Intro

Machine Learning (ML) involves algorithms that enhance performance on tasks through experience, represented as < P, T, E >. Key applications include pattern recognition, anomaly detection, and prediction, with various learning types such as supervised, unsupervised, and reinforcement learning. The design of an ML system includes data collection, function representation, and algorithm selection to improve task performance.

Uploaded by

Sairam Manne
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
8 views18 pages

1 Intro

Machine Learning (ML) involves algorithms that enhance performance on tasks through experience, represented as < P, T, E >. Key applications include pattern recognition, anomaly detection, and prediction, with various learning types such as supervised, unsupervised, and reinforcement learning. The design of an ML system includes data collection, function representation, and algorithm selection to improve task performance.

Uploaded by

Sairam Manne
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/ 18

Machine Learning

Introduction
• Machine Learning (ML) is the study of algorithms that
improve their performance P, at some task T , with experience
E.
• A learning task is given by < P, T , E >.
• Learning is any process by which a system improves
performance from experience.
• Traditional programming:

1
Introduction (contd.)

• Machine Learning:

2
Example ML tasks
• Pattern Recognition. Eg: Optical Character Recognition
(OCR), Handwriting Recognition, Biometric Recognition,
Speech Recognition, Face Recognition and Medical Imaging.
• Anamoly detection. Eg: Abnormal credit card transactions,
Abnormal patterns of sensor readings in a nuclear power plant.
• Prediction. Eg: Future stock prices or currency exchange
rates.
• State of the art applications – Web search, Information
Retrieval, Remote Sensing, Computational Biology, Robotics,
Social Networks etc.

3
Example ML tasks (contd.)
• ML is required for complex tasks. For example, handwriting
digit recognition process has to consider different writers and
the corresponding writing styles. In the below figure, it is
difficult to recognize digit 2 (even for a human being).

4
Design of an ML system

• ML system gains experience or knowledge on a particular task


from data.
• Collect the training data (experience).

5
Design of an ML system (contd.)

• Choose the target function (that means, task to be learnt).


• Represent the target function (Eg: Linear Regression, Neural
Networks, Support Vector Machines etc.).
• Choose a learning algorithm to infer the target function from
the data (experience). Eg: Gradient descent, Genetic
algorithms etc.

6
Design of an ML system (contd.)

• Loop over the following steps:


– Understand problem domain, prior knowledge, and goals.
– Data collection, pre-processing and feature extraction.
– Learn models (target function).
– Evaluate and interpret results.
– Consolidate and deploy discovered knowledge.

7
Components of ML algorithm

• Representation of target function. Eg: Linear Regression,


Logistic Regression, Neural Networks, Support Vector
machines, Decision Trees, Nearest Neighbour, Hidden Markov
Models (HMMs)etc.
• Optimization of function. Eg: Gradient descent (Perceptron
or Backpropagation), Divide and conquer (Decision tree
learning), Dynamic programming (HMM learning),
Evolutionary computation (Genetic algorithms).
• Evaluation of learnt function – Use measures like Accuracy,
Precision, Recall etc.

8
Types of Learning
• Supervised (Inductive) learning –
Given: Training data + desired outputs (labels).
• Unsupervised learning –
Given: Training data (No desired outputs).
• Semi-supervised learning –
Given: Training data + some desired outputs.
• Reinforcement learning –
Rewards for the sequence of actions.

9
Types of Learning (contd.)

• Supervised learning based methods: Linear regression,


Classification using Logistic regression, Neural networks,
Support vector machines, Decision tree learning and Deep
learning.
• Unsupervised learning based methods – Clustering and
Dimensionality reduction.
• Reinforcement learning based methods – Q learning and
Temporal difference learning.

10
Supervised learning: Regression

• Given training data points: (x1 , y1 ), (x2 , y2 ), · · · (xn , yn ).


• Learn a function y = f (x), to predict y for a given x.
• If y is real valued, this learning problem is Regression.

11
Supervised learning: Classification
• Given training data points: (x1 , y1 ), (x2 , y2 ), · · · (xn , yn ).
• Learn a function y = f (x), to predict y for a given x.
• If y is categorical (discrete), this learning problem is
Classification.
• For example, Cancer = f (TumorSize). Here, Cancer is
discrete with two values (Benign or Malignant), a two class
classification based on Tumorsize. Of course, there can be
errors in the classification.

Here, Malignant – Black points. Benign – Gray points.

12
Supervised learning: Classification (contd.)

• In general, each data point xi , can be multi-dimensional.


• For example, Cancer = f (Tumorsize, Age). Here also, Cancer
is discrete with two values (Benign or Malignant), a two class
classification based on two input parameters,
(Tumorsize, Age).

13
Supervised learning: Classification (contd.)

• If y has k discrete values, it is a k-class classification problem.


For example, in Handwritten digit recognition problem, there
are 10 classes (0 – 9).

14
Unsupervised learning: Clustering

• Given data points: x1 , x2 , · · · xn (no labels).


• Learn hidden structure among x values – Clustering.
• For example, clustering of 2D data points (that means, each
xi is a 2D point) is shown below:

• In general, each data point xi , can be multi-dimensional.

15
Reinforcement learning

• Rewards for actions.


• Given a sequence of states and actions with rewards, learn an
optimal policy.
• Policy is a mapping from states to actions.
• Examples: Robot navigation, Autonomous driving etc.

16
Thank You

17

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