arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2107.04562stat.MLcs.LG
收藏

贝叶斯学习规则

The Bayesian Learning Rule

RIKEN Center for Advanced Intelligence Project(理化学研究所先进智能项目中心);King Abdullah University of Science and Technology(阿卜杜拉国王科技大学)

Mohammad Emtiyaz Khan, Håvard Rue

AI总结:

本文提出贝叶斯学习规则,用自然梯度估计候选分布来近似后验,从而统一并推广了从岭回归、牛顿法到随机梯度下降、Dropout 等众多经典与现代机器学习算法,并支持设计新算法。

摘要:

我们证明许多机器学习算法都是单一算法——贝叶斯学习规则——的具体实例。该规则源自贝叶斯原理,可产生来自优化、深度学习和图模型等领域的广泛算法。这包括经典算法,如岭回归、牛顿法和卡尔曼滤波,以及现代深度学习算法,如随机梯度下降、RMSprop 和 Dropout。推导此类算法的关键思想是使用通过自然梯度估计的候选分布来近似后验分布。不同的候选分布会产生不同的算法,而对自然梯度的进一步近似则会产生这些算法的变体。我们的工作不仅统一、推广并改进了现有算法,还帮助我们设计新算法。

英文摘要:

We show that many machine-learning algorithms are specific instances of a single algorithm called the \emph{Bayesian learning rule}. The rule, derived from Bayesian principles, yields a wide-range of algorithms from fields such as optimization, deep learning, and graphical models. This includes classical algorithms such as ridge regression, Newton's method, and Kalman filter, as well as modern deep-learning algorithms such as stochastic-gradient descent, RMSprop, and Dropout. The key idea in deriving such algorithms is to approximate the posterior using candidate distributions estimated by using natural gradients. Different candidate distributions result in different algorithms and further approximations to natural gradients give rise to variants of those algorithms. Our work not only unifies, generalizes, and improves existing algorithms, but also helps us design new ones.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑