Reproduce “Beyond backpropagation: bilevel optimization through implicit differentiation and equilibrium propagation” 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
Nicolas Zucchet, João Sacramento
The authors review bilevel learning: an inner optimization produces a state, and an outer loss evaluates it. They explain implicit differentiation, which estimates the outer gradient through an auxiliary solve, and equilibrium propagation, which compares derivatives at unnudged and nudged solutions. Under stated assumptions, implicit differentiation has a gradient-error bound linear in approximation errors; the comparison is theoretical, not empirical.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.