PAC(m)-Bayes Narrowing the Empirical Risk Gap in the Misspecified Bayesian Regime
INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS, VOL 151(2022)
摘要
The Bayesian posterior minimizes the "inferential risk" which itself bounds the "predictive risk." This bound is tight when the likelihood and prior are well-specified. However since misspecification induces a gap, the Bayesian posterior predictive distribution may have poor generalization performance. This work develops a multi-sample loss (PAC(m)) which can close the gap by spanning a trade-off between the two risks. The loss is computationally favorable and offers PAC generalization guarantees. Empirical study demonstrates improvement to the predictive distribution.
更多查看译文
关键词
empirical risk gap,regime
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络