Reproduce “Score-based generative models learn manifold-like structures with constrained mixing” 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
Li Kevin Wenliang, Ben Moran
The authors analyze local geometry in a score model trained on MNIST. They find that some input and output directions are misaligned, while their subspaces substantially overlap with an effective conservative field’s data-associated subspace. Using 128 generated images and a dimensionality cutoff with κ set to 0.1, they find estimated local dimensionality rises and varies more as noise decreases, in this model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.