Reproduce “ISAAC Newton: Input-based Approximate Curvature for Newton's Method” 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

Felix Petersen, Tobias Sutter, Christian Borgelt, Dongsung Huh, Hilde Kuehne, Yuekai Sun, Oliver Deussen

ISAAC uses a limiting curvature-based update to condition gradients with layer inputs, avoiding an output-gradient factor used by K-FAC. Its added computation is asymptotically small relative to gradient computation when batch size is much smaller than layer width. In a reported MNIST autoencoder setting, it reaches low loss sooner than K-FAC in elapsed time, but is unstable at regularization 0.01.

Summary by alphaXiv.

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