Reproduce “Extendable and invertible manifold learning with geometry regularized autoencoders” 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

Andrés F. Duque, Sacha Morin, Guy Wolf, Kevin R. Moon

GRAE (geometry-regularized autoencoder) trains an encoder and decoder while softly matching encoder codes to a precomputed PHATE embedding. On the Swiss Roll test, its two-dimensional embedding predicted the known factor with R² of 0.9762, versus 0.5136 for a vanilla autoencoder, and reconstruction MSE was 0.0034 versus 0.0210. Results support gains on tested tasks, not universal superiority.

Summary by alphaXiv.

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