Reproduce “Using local plasticity rules to train recurrent neural 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
Owen Marschall, Kyunghyun Cho, Cristina Savin
The authors propose a recurrent-network model that uses three synapse populations and two activity phases to approximate temporal credit assignment with local plasticity rules. W runs the task, while A and J support learning. On the (4, 6)-back task, the model reaches 75% accuracy, the task’s ceiling, learns the 4-step dependency, and shows only partial evidence of learning the 6-step dependency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.