Reproduce “The Information Geometry of Representation Formation” 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
Sanjukta Krishnagopal
The authors track how a network’s hidden representation changes across training checkpoints using Bures tortuosity, the ratio of checkpoint travel to endpoint distance. In binary parity tasks with fixed two-class outputs, mean tortuosity rose from 1.014 at interaction order 1 to 4.177 at order 4. The controls support this ordering, but do not establish a universal pattern.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.