Reproduce “Factorized Gradients for Scalable Highly-expressive Parametric Diffeomorphisms” 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
Amit Aflalo, Eran Treister, Chaim Baskin, Oren Freifeld
FG-CPAB reduces CPAB gradient cost by collecting trajectory-dependent information per cell, then projecting it onto the parameter basis. This changes backward-computation scaling from O(dTN) to O(TN + dC). In a 2D benchmark against libcpab, it reaches 10,463× gradient speedup and 509× lower peak GPU memory, while scaling to tessellation size 30; benchmark results, not end-to-end fitting speedups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.