This paper explores quantitative Laplace-type convergence results for exponential probability measures, focusing on norm-like potentials. It establishes bounds between measures $\pi_\varepsilon$ and $\pi_0$ using Wasserstein distance under a generalized Jacobian invertibility condition. The research utilizes geometric measure theory tools and applies the findings to maximum entropy models and the convergence of Stochastic Gradient Langevin Dynamics at low temperatures for non-convex minimization. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Provides theoretical underpinnings for understanding the behavior of optimization algorithms like SGLD in low-temperature, non-convex settings.
RANK_REASON This is a research paper published on arXiv concerning mathematical probability and its application to algorithms.