Reproduce “Inductive Bias, Not Physics: What Neural Renormalization-Group and Machine-Learned Holographic Maps Can and Cannot Identify” 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
Yan Ru Pei
The paper examines whether learned renormalization-group methods identify coarse physical quantities from fine lattice data. It shows smooth bijections can exactly match factorized targets at every admissible coarse coupling, so fit alone does not identify physics. In a Gaussian model, general linear maps were exact at six tested masses; finite-budget RealNVP trials favored one, but were unconverged.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.