Reproduce “Dynamic Manifold Hopfield Networks for Context-Dependent Associative Memory” 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
Chong Li, Taiping Zeng, Xiangyang Xue, Jianfeng Feng
Dynamic Manifold Hopfield Networks recover patterns from noisy or incomplete cues by using each cue to initialize activity and condition weights and biases, creating a cue-dependent retrieval landscape. In an N-unit network storing 2N patterns, they average 64% accuracy versus 1% for Classical and 13% for Modern Hopfield Networks; on balanced patterns, they score 25% versus 26% for Modern Hopfield Networks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.