A Limitation of the PAC-Bayes Framework
NIPS 2020, 2020.
Theorem 2 does not exclude the possibility of Probably Approximately Correct-learning thresholds over Xn with sample complexity that scale with O(log∗ n) such that the PAC-Bayes bound vanishes
PAC-Bayes is a useful framework for deriving generalization bounds which was introduced by McAllester ('98). This framework has the flexibility of deriving distribution- and algorithm-dependent bounds, which are often tighter than VC-related uniform convergence bounds. In this manuscript we present a limitation for the PAC-Bayes framewo...More
PPT (Upload PPT)