Reproduce “The Geometry of Meaning in Neural Manifolds” 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
Valentino Maiorca
The authors examine how semantic correspondences can align independently trained neural-network representations for model stitching. Their latent functional map method uses graph-Laplacian eigenbases to expand a small set of known matches, then fits an orthogonal transformation between native latent spaces. On CIFAR-100 coarse classification, it reaches 0.28 accuracy with 5 anchors, versus 0.08 for direct transformation; this advantage is not universal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.