Hi, I'm Hooman Zolfaghari

Understand at a precise mathematical level what capabilities and information are fundamentally needed to learn different kinds of tasks successfully, and to understand the basic algorithmic principles involved in getting computers to learn from data and to improve performance with feedback (ref).
Proving guarantees for algorithms (under what conditions will they succeed ?, how much data and computation time is needed ?) and developing machine learning algorithms that provably meet desired criteria.
Particularly researching problems in Sample Comlexity, Learnability, Generalization, Stability, Kernels, or more specific problems of Deep Learning Theory, Domain Adaptation, Mathematics of Attention, Nueral Tangent Kernel
Consquently, I deeply enjoy topics related to High-Dimensional Probability and Statistics.

geometric unification of a broad class of ML problems from the perspectives of symmetry and invariance (ref). Extending models and theories to use non-Euclidean data, such as graphs. Learn and generalize from structured data by incorporating geometric priors into the models.