Introduction

CS771: Introduction to Machine Learning

Purushottam Kar

CS771: Intro to ML
Course Details
Name: CS771(A) Introduction to Machine Learning
Nickname: ML
Lectures: Wednesdays and Fridays, 1800-1930 hrs, L16
Instructor: Purushottam Puru Kar
Teaching assistants: wait for next slide
Course websites
Internal website: http://tinyurl.com/ml17-18ai
Piazza website: http://tinyurl.com/ml17-18adf (Enroll !!)
Office hours:
Puru: Thursdays 1800-1930 hrs, RM509 (by appointment)
TA hours to be announced
August 2, 2017 2
CS771: Intro to ML
Course Mentors

Jayant Agrawal (agjayant) Nishit Asnani (nishit)

Computer vision, ML NLP, Reinforcement learning

Ankita Bishnu (ankitab)

Shibhansh Dohare (sdohare)
Computer vision and NLP,
NLP, Reinforcement learning
VQA

August 2, 2017 3
CS771: Intro to ML
Course Mentors
Abhinav Garg (abhigarg)
Atul Gangwar (atulgang)
NLP, Computer vision,VQA,
Information retrieval
Graph algorithms

Govind Gopakumar (govindg) Purushottam Kar (purushot)

Optimization, ML Online learning, Optimization

August 2, 2017 4
CS771: Intro to ML
Course Mentors
Rohit Singh Kharanghar (rsk)
Gopichand Kotana (gopick)
Computer vision, Deep
Computer vision, ML
learning

Bhaskar Pratim Mukhoty

Pawan Kumar (kpawan)
(bhaskarm)
Deep learning, NLP, Machine
Adversarial ML, Non-convex
translation
optimization

August 2, 2017 5
CS771: Intro to ML
Course Mentors
Vishak Prasad (vishak) Munot Rushab Preetam
(rushab)
Bayesian ML, Deep learning,
Visual recognition NLP, Learning with kernels

Arindam Sarkar (arindam) Susmit Wagle (waglesmi)

Probabilistic ML, Computer ML, Computer vision, NLP,

vision Computer security

August 2, 2017 6
CS771: Intro to ML
Auditors
Please send a mail to Puru to be included in mailing list
Auditors will have access to several aspects of the course
Lectures and lecture material
Discussion forum activities
Assignment and examination questions
Project presentations
We regret our inability to extend the following services
Submit assignments and receive graded submissions
Appear for examinations and receive graded answer scripts
Take part in project groups with registered students
If few auditors wish to form a separate project group, talk to Puru

August 2, 2017 7
CS771: Intro to ML
Grading Scheme
Assignments 30%
Term project 20%
End semester examination - 30%
Mid semester examination - 20%

Excellent work in term project will be very well rewarded*

August 2, 2017 8
CS771: Intro to ML
Assignments
Pen-paper type
Reasoning about ML algorithms
Developing extensions to ML algorithms
Derivations and calculations
Must be submitted electronically (details later)
Must be typeset in LaTeX using style file provided
Several resources available online start learning today!
Scanned/photocopied submissions will be rejected without review
Use Lyx with caution (must confirm to style file)
Use other typesetting software at your own risk

August 2, 2017 9
CS771: Intro to ML
Assignments
Programming type
Develop basic ML primitives
Use toolboxes to solve problems
Develop extensions
All submissions will be done electronically (details later)
Language of choice: Python
Several resources available online start learning today!
Plagiarism checks will be carried out renaming identifiers is futile!
Code as well as learnt models will have to be submitted

August 2, 2017 10
CS771: Intro to ML
Term Project
Form groups of 4 or 5 start planning today!
Larger or smaller groups need to justify
Groups must be formed by August 15, 2017 intimate Puru
Several project ideas will be floated you can propose your own
Talk with your friends, discuss with course mentors
Look around you for inspiration ML can solve a lot of problems
Focus on getting motivated with the problem solutions will come later
A formal project proposal will be due on September 15, 2017
Pinpoint the problem you wish to investigate
Give a description of existing work on that problem
Give a rough outline on what plan of action you wish to take

August 2, 2017 11
CS771: Intro to ML
Reference Material
No textbook for the course
Reference material will be put up on internal website
Locally cached copies for some references

August 2, 2017 12
CS771: Intro to ML
Use of Unfair Means
The following are prohibited severe penalties
Copying answers in pen-paper assignments
Copying code in programming assignments
Passing off known results as ones own
Manipulating experimental results

The following are prohibited credit deductions

Using material in project report (figures, text) without acknowledging
Using help from auditors or mentors in projects without acknowledging

August 2, 2017 13
CS771: Intro to ML
 What is Machine Learning?
and where is it used?

August 2, 2017 14
CS771: Intro to ML
Machine Learning
The art and science of designing adaptive algorithms

August 2, 2017 15
CS771: Intro to ML
Machine Learning
The art and science of designing adaptive algorithms

August 2, 2017 16
CS771: Intro to ML
How to identify ML applications
Complexity: no closed form solutions
Human experts cannot successfully prescribe a solution
Graph cut problem not a good candidate for ML

Presence of immense variety

Too many variants to be solved independently
Recommendation systems excellent ML problems

Need for automation

Scalability and speed are main criterion
Do we need to automate medicine, driving?

August 2, 2017 17
CS771: Intro to ML
Prominent ML applications

August 2, 2017 18
CS771: Intro to ML
 How is ML done?
the ultra short version

August 2, 2017 19
CS771: Intro to ML
A typical exam prep cycle
TEST TEST TEST

Preparation
Practice Tests
Actual Exam
EXAM

Subject
Pass / Fail Knowledge
August 2, 2017 20
CS771: Intro to ML
 Spam Filtering

August 2, 2017 21
CS771: Intro to ML
A typical ML workflow

ML
Algorithm
Existing Emails

New Emails

Spam
Spam/Non-spam Filter
August 2, 2017 22
CS771: Intro to ML
A typical ML workflow

ML
Algorithm
Training Data

Test Data

Model
Output
August 2, 2017 23
CS771: Intro to ML
ML as an examination
EXAM

Critical to do well on D-day Critical to do well on test data

Prep test results indicative Training accuracies indicative
No out-of syllabus questions Training/test data are similar
Should not leak exam paper Should not look at test data

Many other similarities future discussions

August 2, 2017 24
CS771: Intro to ML
Input driven ML

Training Data

Batch Learning Active Learning

Robust Learning
Semi-supervised Supervised Learning
Learning
Online Learning
Unsupervised Learning Reinforcement Learning

August 2, 2017 26
CS771: Intro to ML
Output driven ML
Test Data

Regression Multi-classification

URGENT,
Topic Modelling Tagging OFFICIAL,
TAX

Ranking
> > > Binary Classification
August 2, 2017 28
CS771: Intro to ML
Process driven ML
ML Algorithm
Local FA Methods
Methods Neural Methods

Kernel Methods Probabilistic Methods

August 2, 2017 30
CS771: Intro to ML