Reproduce “Generative Replay for Equilibrium Propagation: Toward Edge-Friendly Continual Learning in 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
Victor Cook, Gilles Sassatelli, Thierry Gil, Mohamed Watfa, Marina Reyboz
The authors introduce an EqProp Hopfield network that generates pseudo-examples for continual learning without stored examples or a separate generator. During training, they release some input units and add a reconstruction objective; later, a frozen copy relaxes from unconstrained states to produce replay data. On Split-CIFAR10, EB-Gen reaches 56.1 ± 0.9% average accuracy, below Experience Replay’s 76.6 ± 0.5%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.