Reproduce “AutoHebb: Self-Modifying Neural Networks that Zero-Shot Infer and Continually Learn” 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
Daniel Levine, David van Dijk, Kenneth Kay
AutoHebb lets a network regulate its own learning to infer relations across new examples while retaining prior knowledge. A controller combines feedback with network activity to gate local Hebbian weight updates; gradient-based meta-learning tunes the system. In uncued TI-to-TE tests, it outperformed ANML overall, but same-group TE accuracy was 0.614 ± 0.032, on controlled synthetic benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.