Reproduce “Mean-Field Assisted Deep Boltzmann Learning with Probabilistic Computers” 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
Shuvro Chowdhury, Shaila Niazi, Kerem Y. Camsari
The authors train Boltzmann machines by estimating data-conditioned positive-phase statistics with mean-field approximations and sampling negative-phase statistics using Gibbs sampling on an FPGA. At epoch 100, correlation errors are about 1% in the positive phase and 55% in the negative phase. On full MNIST, accuracy is about 87% with Gibbs sampling and 70% with naive MFT; HMFT was not tested there.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.