Arora group:  

Learning algorithms with performance guarantees through data-structural/distributional assumptions, unsupervised learning, language models (including topic models and word embeddings), matrix and tensor factorization, deep nets, sparse coding.

Hazan group:

Efficient optimization algorithms for machine learning,  non-generative unsupervised and semi-supervised learning, online convex optimization, and regret minimization in games.

Lee group:


Jin group:

Reinforcement Learning, game theory, optimization, representation/meta-learning, and statistics.