Reproduce “Bridging the Gaps Between Residual Learning, Recurrent Neural Networks and Visual Cortex” 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

Qianli Liao, Tomaso Poggio

The authors show residual networks can be recurrent when layers share weights, reusing one learned transformation over time. This equivalence applies to a time-invariant case, not ordinary untied ResNets. On CIFAR-10, a shared two-state model uses 297,994 parameters versus 2,365,962 without sharing; errors are similar in range, but the unshared model finishes lower. This parameter tradeoff does not establish a cortical mechanism.

Summary by alphaXiv.

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