Reproduce “AlgebraNets” 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

Jordan Hoffmann, Simon Schmitt, Simon Osindero, Karen Simonyan, Erich Elsen

AlgebraNets replace scalar weights and activations in neural networks with structured tuples whose components interact through alternative multiplication rules. On ImageNet, the authors report that the 2×2 real-matrix algebra M₂(ℝ) is their most parameter-efficient tested choice, without exceeding real-baseline FLOPs in ResNet-50 or MobileNet-v1. On Enwik8, it matches validation BPC with fewer parameters but higher reported FLOPs.

Summary by alphaXiv.

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