collaborators

5 papers

cs.DS2024

Efficient Discrepancy Testing for Learning with Distribution Shift

Gautam Chandrasekaran, Adam R. Klivans, Vasilis Kontonis +2

A fundamental notion of distance between train and test distributions from the field of domain adaptation is discrepancy distance. While in general hard to compute, here we provide…

cs.DS2024

Tolerant Algorithms for Learning with Arbitrary Covariate Shift

Surbhi Goel, Abhishek Shetty, Konstantinos Stavropoulos +1

We study the problem of learning under arbitrary distribution shift, where the learner is trained on a labeled set from one distribution but evaluated on a different, potentially a…

cs.DS2023

Agnostic proper learning of monotone functions: beyond the black-box correction barrier

Jane Lange, Arsen Vasilyan

We give the first agnostic, efficient, proper learning algorithm for monotone Boolean functions. Given uniformly random examples of an unknown…

cs.LG2023

Tester-Learners for Halfspaces: Universal Algorithms

Aravind Gollakota, Adam R. Klivans, Konstantinos Stavropoulos +1

We give the first tester-learner for halfspaces that succeeds universally over a wide class of structured distributions. Our universal tester-learner runs in fully polynomial time…

cs.LG2023

An Efficient Tester-Learner for Halfspaces

Aravind Gollakota, Adam R. Klivans, Konstantinos Stavropoulos +1

We give the first efficient algorithm for learning halfspaces in the testable learning model recently defined by Rubinfeld and Vasilyan (2023). In this model, a learner certifies t…