Reproduce “Center-Manifold Reduction of Learning at Bifurcations: Interference and Rich Learning in 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
James Hazelden, Eric Shea-Brown
The authors model how recurrent neural network learning changes near dynamical bifurcations by approximating the parameter-to-state Jacobian with a low-rank channel tied to the critical state-space direction. In controlled 32-dimensional networks, the pitchfork case reached effective rank about 1.04 and 0.7% relative Jacobian error near the event. The approximation is local and conditional, and can explain selective task interference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.