BriefGPT.xyz
May, 2023
受限领域中的无学习率贝叶斯推理
Learning Rate Free Bayesian Inference in Constrained Domains
HTML
PDF
Louis Sharrock, Lester Mackey, Christopher Nemeth
TL;DR
通过利用凸优化中的coin betting思想,将约束取样视为概率测度空间上的镜像优化问题,我们介绍了一套完全不依赖学习率的基于粒子的约束取样算法,并引入了现有约束取样算法的统一框架。数值实验表明,我们的算法在清单、公平约束、后选推理等任务中实现了非常有竞争力的性能,且无需调节任何超参数。
Abstract
We introduce a suite of new particle-based algorithms for sampling on constrained domains which are entirely learning rate free. Our approach leverages
coin betting
ideas from convex optimisation, and the viewpoint of
c
→