Search papers, labs, and topics across Lattice.
This paper introduces the Gradient-free Riemannian Langevin Sampler (GRiLS), a novel method designed to efficiently sample from multimodal probability distributions without relying on gradient evaluations. By employing a Riemannian metric to reshape local geometry, GRiLS enhances exploration and facilitates transitions between modes, addressing the common issues of poor mixing and mode trapping seen in traditional Markov Chain Monte Carlo methods. Empirical evaluations on multimodal benchmarks reveal that GRiLS significantly outperforms both gradient-based and existing gradient-free MCMC techniques in terms of mixing efficiency.
GRiLS achieves superior mixing in multimodal distributions without the need for gradients, transforming the landscape of sampling methods.
We address the problem of efficiently sampling multimodal probability distributions, where standard Markov Chain Monte Carlo methods often suffer from poor mixing and mode trapping. To mitigate these issues, we propose Gradient-free Riemannian Langevin Sampler (GRiLS), a novel proposal that improves exploration without requiring gradient evaluations of the target density. Our approach introduces a Riemannian metric which reshapes the local geometry in order to facilitate transitions across modes. The resulting gradient-free MCMC algorithm is particularly suitable for complex, computationally expensive targets where derivatives are unavailable or impractical. The GRiLS proposal requires knowing the mean and covariance of the target density, which we estimate using an ensemble of interacting particles. Empirical results on multimodal benchmarks demonstrate that GRiLS achieves improved mixing compared to existing gradient-based and gradient-free MCMC approaches.
