Fall 2024 (Aug-Dec)

Online Prediction and Learning

Course No.: E1 245

Instructors: Aditya Gopalan (firstname AT iisc.ac.in) and Rahul Singh (fullname AT iisc.ac.in), Dept. of ECE

Time: T,Th 10:00-11:30 (First meeting on Tue 6 Aug 2024)

Place: MP 30 (ECE dept.)
Class participants will need to join a Teams portal for this course (for announcements, discussions and course material); Click here to join the class Team.

Course Description: Online learning is a class of sequential-decision making problems where the agent faces uncertainty about the future. These problems arise within many modern data-driven intelligent systems (think Internet recommendation engines, financial portfolio allocation, resource allocation in communication systems, etc.) where one needs the ability to continuously and efficiently adapt to changing conditions. This course exposes students to common formulations and algorithms for regret minimization and sequential inference, while introducing tools for rigorous performance analysis along the way.

Contents: Online convex optimization — Follow the leader, Follow the regularized leader, Online mirror descent, Learning with expert advice, Adversarial bandits and EXP3; Multi-armed bandits — Upper Confidence Bound, Thompson sampling, Linear bandits, Contextual bandits.

Prerequisites: A course in probability or random processes, and basic knowledge of multivariable calculus (i.e., familiarity with gradients of functions of several variables, Taylor series). Exposure to convexity (convex geometry, convex analysis or convex optimization) will be helpful but is not absolutely necessary. Contact the instructor for clarifications.

References: The course will be based on material from the following texts:

Last updated: 28-Jul-2024, 23:23:54 IST