The Progressive Tempering Sampler with Diffusion (PTSD) is proposed as a solution to enhance the efficiency of sampling from unnormalized densities by combining the strengths of Parallel Tempering (PT) and sequentially trained diffusion models. PTSD generates uncorrelated samples across temperature levels while enabling efficient reuse of sample information, significantly improving target evaluation efficiency compared to traditional diffusion-based neural samplers.
machine-learning ✓
sampling ✓
diffusion ✓
parallel-tempering ✓
+ efficiency