Reproduce “Unipolar Neuron Networks Trained by Recirculation for Reward-Based Continual Learning” 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
Florin Leon
The authors introduce a continual-learning system with branching neurons, task-specific branch inhibition, and reward-based recirculation that sends requested changes back through the same connections. On five two-digit MNIST tasks, with task identity supplied during training and testing, recirculation achieved mean final accuracy 0.939 and forgetting 0.009. Without the task code, final accuracy fell to 0.195 and forgetting rose to 0.875.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.