PulseAugur
LIVE 11:41:03
tool · [1 source] ·
31
tool

New sampling method uses transformers for Gibbs distributions

Researchers have developed a new sampling method for Gibbs distributions using a finite number of particles. This technique, termed preconditioned regularized Wasserstein proximal sampling, approximates the score function with a regularized Wasserstein proximal operator. The method's diffusion component can be interpreted as a modified self-attention mechanism, similar to those in transformer architectures. Experiments show improved stability and performance in various applications, including Bayesian neural network training. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel sampling technique with potential applications in Bayesian neural network training and image deconvolution.

RANK_REASON The cluster contains a new academic paper detailing a novel sampling method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

COVERAGE [1]

  1. arXiv stat.ML TIER_1 · Hong Ye Tan, Stanley Osher, Wuchen Li ·

    Preconditioned Regularized Wasserstein Proximal Sampling

    arXiv:2509.01685v2 Announce Type: replace Abstract: We consider sampling from a Gibbs distribution by evolving finitely many particles. We propose a preconditioned version of a recently proposed noise-free sampling method, governed by approximating the score function with the num…