<ul><li>Recent research has focused on designing neural samplers to sample from unnormalized densities efficiently.</li><li>Progressive Tempering Sampler with Diffusion (PTSD) is proposed to improve the efficiency of target evaluations by training diffusion models sequentially across temperatures.</li><li>PTSD leverages the advantages of Parallel Tempering (PT) to enhance the training of neural samplers and generate well-mixed, uncorrelated samples.</li><li>The method significantly boosts target evaluation efficiency, outperforming diffusion-based neural samplers.</li></ul>

Progressive Tempering Sampler with Diffusion

Discover more