Reproduce “Attention in a family of Boltzmann machines emerging from modern Hopfield networks” 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
Toshihiro Ota, Ryo Karakida
AttnBM turns the attention-derived energy into a Boltzmann probability model and enables exact likelihood training when inverse temperature β is a positive integer. At β=1, its distribution is a Gaussian mixture centered on memory vectors, and the authors derive exact gradients. In toy image experiments, it reconstructs training examples at P=200, but not at P=50,000, without establishing held-out generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.