Poincar\'e Inequality for Local Log-Polyak-Lojasiewicz Measures : Non-asymptotic Analysis in Low-temperature Regime
Yun Gong, Zebang Shen, Niao He

TL;DR
This paper analyzes the convergence of Langevin dynamics in complex non-convex landscapes with connected local minima, establishing an pendent Poincare9 inequality and convergence rate in the low-temperature regime.
Contribution
It introduces a class of measures with local Polyak-a7ojasiewicz conditions and connected minima, providing non-asymptotic convergence analysis in non-convex settings.
Findings
Poincare9 constant lower bounded independently of psilon
Langevin dynamics converges at rate b0(1/psilon) in low-temperature regime
Potential functions can have non-isolated minima forming a connected manifold
Abstract
Potential functions in highly pertinent applications, such as deep learning in over-parameterized regime, are empirically observed to admit non-isolated minima. To understand the convergence behavior of stochastic dynamics in such landscapes, we propose to study the class of \logPLmeasure\ measures , where the potential satisfies a local Polyak-{\L}ojasiewicz (P\L) inequality, and its set of local minima is provably \emph{connected}. Notably, potentials in this class can exhibit local maxima and we characterize its optimal set S to be a compact \emph{embedding submanifold} of without boundary. The \emph{non-contractibility} of S distinguishes our function class from the classical convex setting topologically. Moreover, the embedding structure induces a naturally defined Laplacian-Beltrami operator on S, and we…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsStochastic processes and statistical mechanics · Spectral Theory in Mathematical Physics · Stochastic processes and financial applications
MethodsDiffusion · Sparse Evolutionary Training
