Reproduce “A novel set of rotationally and translationally invariant features for images based on the non-commutative bispectrum” 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
Risi Kondor
The authors introduce a bispectrum representation for recognizing handwritten digits despite translation and rotation. They map monochrome images onto a sphere and construct features invariant to sphere rotations, then compare SVMs using these features with raw-pixel baselines. On transformed NIST digits, the 8-versus-9 RBF classification error is 10.06% versus 44.06%. This is a small-sample pairwise result; invariance is limited by projection and bandwidth errors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.