Reproduce “Training dynamically balanced excitatory-inhibitory 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
Alessandro Ingrosso, L.F. Abbott
The authors train excitatory–inhibitory networks to perform tasks while obeying Dale’s law, using target-based learning and bounded constrained coordinate descent to fit recurrent connections. They find that currents and ordinary L2 regularization can produce parametric balance, with large currents canceling. In setups, balanced initialization and a penalty on weight changes preserve dynamic balance; task results include 96% temporal XOR accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.