COL870/8385: Special Topics in Machine Learning

Instructor: Adarsh Barik
Credit: 3 (3-0-0)
Semester 2 (2025-2026)
TF 3:30-5 PM

Optimization for Machine Learning

COL870/COL8385 is a 3-credit Special Topics course in Machine Learning. The course will cover topics in optimization in both offline and online settings. The material will be motivated throughout by applications to modern machine learning problems, and will include both foundational ideas and advanced topics.

Expected background

This course is intended for both (post)graduate and undergraduate students interested in the optimization foundations of machine learning. A basic level of mathematical maturity is expected; students with concerns about their background should consult the instructor.

A fundamental understanding of linear algebra, calculus, and probability theory is required for this course. Prior experience or familiarity with machine learning is also beneficial, though not mandatory.

Learning outcomes

Upon completing this course, students will be familiar with key concepts in optimization, enabling them to:

  1. Understand the fundamental concepts such as convex sets, convex functions and optimality criteria for the optimization problems

  2. Understand first-order optimization algorithms and analyze their convergence properties

  3. Gain familiarity with the foundational concepts in online learning

Content

Key topics include convex sets and functions, conjugates, subdifferentials, primal and dual problem formulations, strong and weak duality, minimax characterizations, and optimality conditions including the Karush-Kuhn-Tucker (KKT) criteria. The course will also cover first-order optimization methods such as gradient descent, stochastic gradient descent (SGD), accelerated gradient techniques, subgradient methods, and Frank-Wolfe algorithms. Additionally, foundational concepts in online learning will be explored, including online gradient descent, online mirror descent, Follow-The-Regularized-Leader (FTRL), and parameter-free algorithms.

Tentative Grading Scheme

Midsemester Exam 30%
Final Exam 30%
Assignments 25%
Scribe 10%
In-class participation 5%

Audit Policy

Minimum 75% attendance and marks equivalent to a grade B or above will be considered as audit pass.

Scribe

In-class participation

Collaboration and use of AI Tools

Reference textbook

  1. Convex Optimization. S. Boyd and L. Vandenberghe. Cambridge University Press, Cambridge, 2003

  2. Introductory Lectures on Convex Optimization: A Basic Course. Y. Nesterov. Kluwer, 2004.

  3. Numerical Optimization. J. Nocedal and S. J. Wright, Springer Series in Operations Research, Springer-Verlag, New York, 2006 (2nd edition).

  4. A Modern Introduction to Online Learning. Francesco Orabona. arXiv preprint arXiv:1912.13213 (2019).