Dimension-free regimes for unadjusted Langevin in ridge-separable models
What it is
This paper analyzes the Unadjusted Langevin Algorithm (ULA) for high-dimensional Bayesian inference using subspace robust Wasserstein distance. It proves that for ridge-separable models, the algorithm achieves dimension-free complexity in overparameterized and large-sample regimes, but not in the proportional regime.
The problem it solves
Classical Wasserstein-2 analyses predict that the bias and mixing time of ULA grow with dimension, making it appear unsuitable for high-dimensional problems despite its practical utility.
What you could use it for
- High-dimensional Bayesian inference where only low-dimensional marginals are of interest
- Analyzing the convergence of MCMC samplers in overparameterized statistical models
Why it matters
Provides theoretical justification for using ULA in high-dimensional settings by showing that dimension-free convergence is possible for specific model classes and regimes, potentially reducing computational costs for Bayesian inference.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.