Reproduce “AntisymmetricRNN: A Dynamical System View on 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

Bo Chang, Minmin Chen, Eldad Haber, Ed H. Chi

The authors introduce AntisymmetricRNN, which uses antisymmetric recurrent feedback and diffusion to stabilize an Euler update while preserving hidden-state information. On noise-padded CIFAR-10, where 32 image rows are followed by 968 Gaussian-noise steps, it reaches 48.3% test accuracy, versus 11.6% for a 128-unit LSTM. On pixel-by-pixel CIFAR-10, the ungated model is roughly on par with the LSTM.

Summary by alphaXiv.

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