Reproduce “Random features for Grassmannian kernel approximation with bounded rank-one projections” 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
Rémi Delogne, Laurent Jacques
The authors develop random features for kernels on fixed-dimensional subspaces, replacing pairwise Gram-matrix comparisons with inner products of compact sketches. Gaussian probes are transformed by sign or periodic maps, inducing different positive-definite kernels; uniform approximation guarantees apply to unstructured constructions. On ETH-80, structured periodic features reached 92.38% accuracy in 5.2383 seconds, versus 92.31% in 86.4142 seconds unstructured; exact projection scored 98.75%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.