You are on page 1of 27

CS5011: Introduction to

Machine Learning
Instructor: B. Ravindran
Slot: R and T
Wed: 2 3:40; Fri: 2 3:40
TAs: Sahil, Sagar, Vamsi, Preksha, Shirjon, Sapana,
Beethika, Ansari
Machine Learning
A Brief Introduction
What is Machine Learning?

said to learn from experience with


respect to some class of tasks, and a
performance measure P, if [the learners]
performance at tasks in the class, as
measured by P, improves with
experience.
Tom Mitchell 1997.

Inductive Learning
3
ML Paradigms
Supervised Learning
Classification misclassification error
Regression squared error
Unsupervised Learning
Clustering purity, spread
Association support, confidence, lift
Reinforcement Learning
Control cost, reward

5
Experience
Training Data

6
Labeled Training Data
Classification

7
Labeled Training Data
Classification
No Buys Computer Yes

Age

Income
eBay Machine Learning: A Brief 8
Introduction
Possible Classifiers

9
Inductive Bias
Need to generalize Assumptions
about lines!
Language bias
In general, Inductive bias
Language bias
Architecture, kernels, feature selection, etc.
Search bias
Maximum margin, cross-validation, etc.

10
The Process
Training Set

X1 , Y 1
X2 , Y 2
Training
X3 , Y 3 Classifier
Algorithm
X4 , Y 4

Test Set Validation


X1 , Y1
X2 , Y2
X3 , Y3
eBay Machine Learning: A Brief 11
Introduction
Training

Input Output
Agent
X
Error Target
Y

12
Applications
Credit Card fraud detection
Valid transaction or not
Network intrusion detection
Valid traffic stream or not
Churn prediction
Potential churner or not
Medical diagnoses
Risk analysis
13
Supervised Learning Architectures
Learning can produce different structures :
Artificial Neural Networks
Support Vector Machines
Decision Trees
Set of Rules
Fuzzy Rules
Bayesian Networks
Nearest Neighbour methods
14
Prediction or Regression

23
Linear Regression
Minimize
sum of squared error

eBay Machine Learning: A Brief 24


Introduction
Applications
Time series predictions
Rainfall in a certain region
Spend on voice calls
Classification!
Data reduction
Trend analysis
Linear or exponential
Risk factor analysis
Factors contributing most to output

25
Labeled Training Data

26
Unlabelled Training Data
Clustering

27
Possible Clusters

28
Applications
Customer Data
Discover classes of customers
Image pixels
Discover regions
Words
Synonyms
Documents
Topics
30
Learning to Control
Familiar models of machine learning
Supervised: Classification, Regression, etc.
Unsupervised: Clustering, Frequent patterns, etc.

How did you learn to cycle?


Neither of the above
Trial and error!
Falling down hurts!

34
RL Framework
Environment

evaluation

State Action
Agent

Learn from close interaction


Stochastic environment
Noisy delayed scalar evaluation
Learn a policy
Maximize a measure of long term performance
35
Applications of RL
Game playing
Backgammon worlds best player!
AlphaGO
Atari Games
Autonomous agents
Robot navigation
Adaptive control
Helicopter pilot!
Combinatorial optimization
VLSI placement
Intelligent Tutoring Systems
36
Autonomous Control

37
Google Deepminds Atari Player

38
Adminisitrivia
No Exams
Flipped Classroom
In Class Tutorials [Best 80% will be taken] (4 marks
each)
There will be 3 review tutorials (10 marks each)
3 math review assignments (3 marks each)
3 programming assignments (30 marks)
One contest (30 marks)

References: PRML Christopher Bishop


Elements of Statistical Learning HTF
Machine Learning Tom Mitchell
More to follow
39
Administrivia
Class structure
30 minutes of discussion
30 minutes for tutorial
20 minutes discussion on tutorial
20 minutes peer grading of the tutorial
Schedule released on Moodle
http://courses.iitm.ac.in/
5011cs2016

40