Reproduce “On Making $\mathrm{SE}(2)$-Invariant Networks Optimal” 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

Tomas Karella, Emily Shinkle, Alice Elisabeth Anastasia Allen, Pieter J. Swart, Nicholas Lubbers, Roxana Bujack

Mom/Inv converts local filter responses into rotation-invariant features using magnitudes and relative phases against a type-one reference. It preserves spatial positions and uses standard pointwise operations, but requires a nonzero reference response. On RotMNIST, it records 99.08±0.04% accuracy, 119,000 parameters, and 0.01 RCI, versus E(2)-CNN’s 98.82±0.01%, 362,000 parameters, and 0.64 RCI. This comparison is limited to that benchmark.

Summary by alphaXiv.

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