Assignments :

  1. Assignment 0: No need to hand it in. [HW0]

  2. Assignment 1: Google form [HW1]

  3. Assignment 2: Out and Graded

Project Topics/Ideas :

  1. Online Learning and Dynamic Systems Or Online Learning With States: Online learning is robust in that it works in adversarial environments and doesn't make any assumptions about data generating process. But the current limitation is that most of online learning methods don't address the question of learning when the system has states or if we are interested in so called policy regret. Some recent works have successfully used results from online learning in Learning under Dynamical systems, control style problems and reinforcement learning. It would be interesting to explore further in this direction as we are just about scratching the surface.

  2. Understanding when/how Neural Networks (Deep learning) generalize: Classical statistical learning theory focus mainly on only model complexity to answer questions of generalization. However much of the success of deep learning can be attributed to the algorithm behind it: Stochastic Gradient Descent. To understand why deep learning works we need to understand algorithmic bias imposed by SGD and other training algorithms in terms of the final model returned, just an analysis of ERM doesn't cut it. Recent papers have made some headway in this problem (well at least for linear networks to begin with).

  3. Using Regret Minimization Strategies in Games: Regret minimizing strategies have always enjoyed close connections to game theory. For zero sum two player games, if both player play regret minimizing strategies, their empirical play converge to minimax equilibria and in more general multiplayer games, players converge to coarse correlated equilibria. Other notions of regret are known to converge to stronger notions of equilibria. Another topic of interest in game theory is to understand quality of play when players all play some specific family of regret minimizing algorithms. There are way many recent and old papers on this topic to explore.

  4. Online Learning and Approximation Algorithms:

  5. Oracle Efficient Contextual Bandits:

  6. General Online Learning With Partial Feedback:

  7. Connections between Learning Theory and Probabilistic Inequalities:

  8. Topics on Computational Learning Theory:

  9. Learning and Differential Privacy:

  10. Theory of Distributed Machine Learning:

  11. Fairness in Machine Learning: