Reproduce “Training Energy-Based Models with Non-MCMC Samplers and Efficient Temperature Estimation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kentaro Kubo, Hayato Goto
The system combines LSB with CEM to estimate sampler temperature, then uses SAL to train SRBMs. In small tests, LSB's mean KL divergence was 0.07 ± 0.01, versus 0.09 ± 0.02 for Gibbs sampling; SAL improved three-spin learning over RBMs. These results are proof of principle; exact Boltzmann sampling and broader applicability remain unproved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.