ELL709
| ELL709 | |
|---|---|
| Online prediction, Optimization, and Learning | |
| Credits | 3 |
| Structure | 3-0-0 |
| Pre-requisites | (MTL106 or ELL711) and ELL706 |
| Overlaps | |
ELL709 : Online prediction, Optimization, and Learning
[edit]Online prediction: Majority and weighted majority algorithms, mistake bound; learning with expert advice-exponential weighted average forecaster, convex loss; non-convex loss, randomization; regret bounds; applications to electrical engineering. Online convex optimization (OCO): Follow the leader (FTL), regularised FTL (FTRL), Online gradient descent (OGD), stochastic gradient descent (SGD), online mirror descent (OMD), regret bounds; applications to electrical engineering. Multi-armed bandit (MAB): Stochastic MAB, UCB algorithm, asymptotic and minimax optimality, KL-UCB algorithm; Adversarial bandits, Exp3 and Exp3-IX algorithms; Contextual and linear bandits; regret analysis; lower bounds; best arm identification with high probability; Bayesian bandit- Thompson sampling, Gittin's index; Restless bandits, Whittle's index; combinatorial and nonstationary bandits; ranking; applications to electrical engineering.