NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:1560
Title:PAC-Bayes under potentially heavy tails

The paper obtains an interesting new PAC-Bayes bound holding for heavy-tailed losses with a logarithmic confidence term by elegantly combining previous concepts. There is currently a lot of interest in the community in such robust estimators, so this result is expected to have a significant impact.