Reproduce “Test-Time Symmetrization: Scaling Laws for Model-Agnostic Learning under Structure” 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

Behrooz Tahmasebi, Melanie Weber

The authors introduce test-time symmetrization for frozen language models by querying equivalent rewritings and aggregating answers. Repeated decoding samples one prompt; symmetrization explores different representations, and a small random sample approximates full group averaging with logarithmic dependence on group size. On 24-node graph edge counting, 16 rewritings achieved 0.92 accuracy versus 0.81 for 16 original-prompt traces, but this does not guarantee accuracy.

Summary by alphaXiv.

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