Reproduce “Global Divergence, Local Convergence: Representation Geometry in SSMs and Transformers” 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
Amit Ben-Artzy, Roy Schwartz
The authors compare residual-stream representations in state-space models and transformers through global geometry, factual-relation decoding, and neighborhood structure. SSM representations are more evenly distributed across dimensions, while transformer representations show greater concentration in the first principal component. Fact-decoding capacity is similar, and local structures align more strongly on factual relations than general text, without establishing identical representations or behavior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.