The authors formalize the distinction between learning at a fixed target error and the standard definition requiring arbitrarily small error.
Every learnable distribution class is also differentially privately learnable
Afzali et al. prove Ashtiani's conjecture for arbitrarily small error, resolving an apparent contradiction from earlier work on fixed-error learning.
Big Tech
Mohammad Afzali · Hassan Ashtiani · Alex Bie · Gautam Kamath · Peihan Liu · Argyris Mouzakis · +1 more
University of British Columbia · McMaster University · Vector Institute · Google Research · New York University
Research Digest··2 min read
Afzali et al.
§