Timezone: »
We present a new concentration of measure inequality for sums of independent bounded random variables, which we name a split-kl inequality. The inequality combines the combinatorial power of the kl inequality with ability to exploit low variance. While for Bernoulli random variables the kl inequality is tighter than the Empirical Bernstein, for random variables taking values inside a bounded interval and having low variance the Empirical Bernstein inequality is tighter than the kl. The proposed split-kl inequality yields the best of both worlds. We discuss an application of the split-kl inequality to bounding excess losses. We also derive a PAC-Bayes-split-kl inequality and use a synthetic example and several UCI datasets to compare it with the PAC-Bayes-kl, PAC-Bayes Empirical Bernstein, PAC-Bayes Unexpected Bernstein, and PAC-Bayes Empirical Bennett inequalities.
Author Information
Yi-Shan Wu (University of Copenhagen)
Yevgeny Seldin (University of Copenhagen)
More from the Same Authors
-
2022 Poster: A Near-Optimal Best-of-Both-Worlds Algorithm for Online Learning with Feedback Graphs »
Chloé Rouyer · Dirk van der Hoeven · Nicolò Cesa-Bianchi · Yevgeny Seldin -
2022 Poster: A Best-of-Both-Worlds Algorithm for Bandits with Delayed Feedback »
Saeed Masoudian · Julian Zimmert · Yevgeny Seldin -
2021 Poster: Chebyshev-Cantelli PAC-Bayes-Bennett Inequality for the Weighted Majority Vote »
Yi-Shan Wu · Andres Masegosa · Stephan Lorenzen · Christian Igel · Yevgeny Seldin -
2020 Poster: Second Order PAC-Bayesian Bounds for the Weighted Majority Vote »
Andres Masegosa · Stephan Lorenzen · Christian Igel · Yevgeny Seldin -
2020 Spotlight: Second Order PAC-Bayesian Bounds for the Weighted Majority Vote »
Andres Masegosa · Stephan Lorenzen · Christian Igel · Yevgeny Seldin -
2019 Poster: Nonstochastic Multiarmed Bandits with Unrestricted Delays »
Tobias Sommer Thune · Nicolò Cesa-Bianchi · Yevgeny Seldin