Reproduce “Natural Gradient Learning in Predictive Coding through Lateral Circuits” 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
Lhea Beumer, Sander de Haan, Niclas Pokel, Pau Vilimelis Aceituno, Benjamin F. Grewe
Natural PC uses a lateral interneuron circuit to estimate presynaptic activity correlations and apply their inverse in local predictive-coding weight updates. On ImageNet-1k reconstruction, it reaches held-out pixel MSE of 1.2 × 10⁻² with 458 times fewer training images and 39 times fewer total training FLOPs than an AdamW autoencoder. Its error later plateaus and worsens, limiting the result to reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.