Reproduce “full-FORCE: A Target-Based Method for Training Recurrent 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

Brian DePasquale, Christopher J. Cueva, Kanaka Rajan, G. Sean Escola, L.F. Abbott

Full-FORCE trains recurrent networks by using recursive least squares to make a task network match the drive of a fixed target network, which receives desired outputs and task inputs during training. Hints can shape that target’s dynamics without being given to the trained network. In modeled tasks, hints improved timing and delayed-comparison performance; results are limited to these simulations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.