Announcements
- We use Canvas for this course; all announcements are made there, so registration is mandatory for students crediting the course. Write to Shubham Gupta (shubhamg [at] iisc.ac.in) if you have not received the invitation (check your spam folder first).
- TA hours: please email a TA for an appointment.
Course description
With the increasing amounts of data being generated in diverse fields such as astronomical sciences, health and life sciences, financial and economic modeling, climate modeling, market analysis, and even defense, there is an increasing need for computational methods that can automatically analyze and learn predictive models from such data.
Machine learning, the study of computer systems and algorithms that automatically improve performance by learning from data, provides such methods; indeed, machine learning techniques are already being used with success in a variety of domains, for example in computer vision to develop face recognition systems, in information retrieval to improve search results, in computational biology to discover new genes, and in drug discovery to prioritize chemical structures for screening.
This course aims to provide a sound introduction to both the theory and practice of machine learning, with the goal of giving students a strong foundation in the subject, enabling them to apply machine learning techniques to real problems, and preparing them for advanced coursework and research in machine learning and related fields.
Prerequisites
E0 232: Probability and Statistics (or an equivalent course elsewhere), with a grade of B or higher. Some background in linear algebra and optimization will also be helpful.
Evaluation (tentative)
- First mid-term examination: February 15
- Second mid-term examination: March 20
Lectures
| # | Date | Topic | Recommended reading |
|---|---|---|---|
| 1 | January 2 | Organisational meeting | |
| 2 | January 4 | Introduction to machine learning; classification using Bayes rule | Duda and Hart: Chapter 2 |
| 3 | January 11 | Introduction to Bayes decision theory | Duda and Hart: Chapter 2 |
| 4 | January 16 | Bayes decision theory (contd.) | Duda and Hart: Chapter 2 |
| 5 | January 18 | Learning as optimization; linear regression | Bishop: Chapter 3 |
| 6 | January 23 | Probabilistic view of linear regression: ML and MAP estimates | |
| 7 | January 30 | Logistic regression: optimization and probabilistic approaches | |
| 8 | February 1 | Parameter estimation for logistic regression: gradient descent, Newton's method, stochastic gradient methods | |
| 9 | February 6 | Hyperplane-based classifiers, the perceptron and the perceptron convergence theorem | |
| 10 | February 9 | Introduction to support vector machines | |
| 11 | February 13 | Support vector machines | |
| 12 | February 20 | Kernel methods and kernel SVM | |
| 13 | February 22 | Feed-forward neural networks and backpropagation | |
| 14 | February 28 | Undirected graphical models, Markov random fields | An Introduction to Restricted Boltzmann Machines |
| 15 | March 4 | Introduction to MCMC and Gibbs sampling | An Introduction to Restricted Boltzmann Machines |
| 16 | March 6 | Restricted Boltzmann machines | An Introduction to Restricted Boltzmann Machines; A Practical Guide to Training Restricted Boltzmann Machines |
| 17 | March 8 | Convolutional and recurrent neural networks | |
| 18 | March 13 | Boltzmann machines and variational methods | Ackley, Hinton and Sejnowski (1985), A Learning Algorithm for Boltzmann Machines; Boltzmann Machines |
| 19 | March 15 | Variational methods (contd.) | Variational methods (see Section 6.3.1) |
| 20 | March 22 | EM algorithm, mixture models and k-means | Bishop: Chapter 9 |
| 21 | March 27 | Bayesian networks; introduction to HMMs | A Tutorial on HMMs |
| 22 | April 2 | HMMs (contd.): the forward–backward algorithm | A Tutorial on HMMs |
Projects
As part of this course, you are required to work on a project. This will give you hands-on experience of working with data from various domains (text, images, videos, etc.) used in contemporary ML research, and also expose you to specialized topics of ML that are more advanced or recent than what will be covered in the lectures. A list of project ideas will be shared with the class; you can also come up with your own project ideas.
Project policy
- Some project ideas are based on reading and surveying several papers; others are based on implementing one or two papers and comparing the proposed methods on datasets. We will point you to relevant papers.
- The best grades will be reserved for groups that come up with new approaches, either theoretical or experimental, that have not been used in the literature.
- Each project will have a project mentor, who will help you locate papers and datasets and clear doubts.
- In implementation-based projects, code may already be available online. It is perfectly fine to use it, as long as it is cited in the reports.
- You are required to submit a mid-term report around the third week of March, and the final report shortly before the final examination.
- Plagiarism is strictly prohibited while writing the reports. Any plagiarism detected will result in an F grade for the course.
Assignment policy
- Assignments are for your learning. You can work in groups, but you must write your own code and solutions (see the academic honesty policy below).
- Late submissions, up to 2 days late (with an advance email to the TA), are allowed with a penalty. After one late submission, no further late submissions will be allowed, except in the case of a documented medical or personal emergency supported by a medical certificate or signed letter submitted to the instructors and to the CSA office.
References
Recommended textbooks
- C. M. Bishop, Pattern Recognition and Machine Learning, Springer, 2006.
- R. O. Duda, P. E. Hart and D. G. Stork, Pattern Classification, Wiley-Interscience, 2nd edition, 2000.
- S. Haykin, Neural Networks: A Comprehensive Foundation, Pearson, 2004.
Additional textbooks
- T. Hastie, R. Tibshirani and J. Friedman, The Elements of Statistical Learning: Data Mining, Inference and Prediction, Springer, 2nd edition, 2009. Free download
- T. Mitchell, Machine Learning, McGraw Hill, 1997.
Optimization textbooks
- R. Fletcher, Practical Methods of Optimization, John Wiley, 2000.
- D. G. Luenberger and Y. Ye, Linear and Nonlinear Programming, Springer.
- Y. Nesterov, Introductory Lectures on Convex Optimization, Springer.
- S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge University Press.
Academic honesty
As students of IISc, we expect you to adhere to the highest standards of academic honesty and integrity.
Elements of the course are designed to support your learning of the subject. Copying will not help you (in the exams or in the real world), so don't do it. If you have difficulties learning some of the topics or lack some background, try to form study groups where you can bounce off ideas with one another and try to teach each other what you understand. You're also welcome to talk to any of us and we'll be glad to help you.
If any exam/assignment is found to be copied, it will automatically result in a zero grade for that assignment and a warning note to your advisor. Any repeat instance will automatically lead to a failing grade in the course.