BriefGPT.xyz
May, 2024
通过反射复制交换随机梯度 Langevin 动力学进行约束探索
Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics
HTML
PDF
Haoyang Zheng, Hengrong Du, Qi Feng, Wei Deng, Guang Lin
TL;DR
利用反射步骤在有界域内进行约束非凸探索，提出了反射式复制交换随机梯度 Langevin 动力学 (r2SGLD)，通过理论和实证研究证明了约束探索对提高模拟效率的关键作用。
Abstract
replica exchange stochastic gradient langevin dynamics
(
resgld
) is an effective sampler for non-convex learning in large-scale datasets. However, the simulation may encounter stagnation issues when the high-tempe
→