Fall 2018 (Aug-Dec)

Online Prediction and Learning

Course No.: E1 245

Instructor: Aditya Gopalan, ECE 2.09, Dept. of ECE, E-mail: first-name AT iisc.ac.in

Time: (TBD)

Place: (TBD)

Course Description: The ability to make continual and accurate forecasts and decisions under uncertainty is key in many of today’s data-driven intelligent systems (e.g. Internet recommendation engines, automated trading, etc.). This elective course is aimed to expose students to techniques for online or sequential learning/decision making under uncertainty. We will explore several frameworks and algorithms for online prediction along with a rigorous understanding of their performance. We will also look at interesting applications of these techniques, such as portfolio optimization (finance), data compression (information theory), etc.

Contents: Online classification; Regret Minimization; Learning with experts; Online convex optimization; Bandits; Applications- sequential investment/portfolio selection, universal lossless data compression, Stochastic games- Blackwell approachability, Learning systems with state- online reinforcement learning

Prerequisites: A basic course in probability or stochastic processes, linear algebra, and (optional) some exposure to convexity or convex optimization. Contact the instructor for clarifications.

Text/References: We will not follow any standard textbook(s). However, a rough basis will be the excellent and encyclopaedic "Prediction, Learning and Games" (PLG). Nicolo Cesa-Bianchi and Gabor Lugosi, Cambridge University Press, 2006 [local PDF copy].

Other useful resources:


Last updated: 2018-04-26, 15:08:18 IST