Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics

arXiv:2405.07839v1 Announce Type: new
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-temperature chain delves too deeply into the distribution tails. To tackle this issue, we propose reflected reSGLD (r2SGLD): an algorithm tailored for constrained non-convex exploration by utilizing reflection steps within a bounded domain. Theoretically, we observe that reducing the diameter of the domain enhances mixing rates, exhibiting a \emph{quadratic} …

abstract arxiv cs.ai cs.lg datasets distribution dynamics exploration gradient however issue replica scale simulation stat.ml stochastic the simulation type via