Researcher, Google
5 papers at NeurIPS 2025
We study a variant of nonuniform PAC learning, where the constants in the learning rate may depend on the marginal distribution, and devise a trichotomy of possible rates.
Giving tight bounds for the acess loss in the Agnostic Learning under Targeted Poisoning modle unsing randomized learnes.
We prove tight asymptotic limits for reconstructing a hidden point from noisy distance queries and give a dimension-based criterion for when metric spaces are (non)-pseudo-finite.
We study learning set functions under one-sided feedback in the PAC learning framework.