Reproduce “A Bochner Weitzenböck Decomposition for Cellular Sheaf Laplacians, Curvature and Neural Sheaf Diffusion” 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
Francesco Maria Guadagnuolo, Alessio Borgi, Dorina Thanou, Pietro Liò
The authors split the sheaf Laplacian into positive-semidefinite Bochner diffusion and block-diagonal Forman terms, making trained operators interpretable. They rank nodes by conviction from learned states and intervene on frozen models. Across eight benchmarks, high-conviction nodes cause larger performance drops than low-conviction nodes at every tested zero-masking budget, supporting an operational ranking rather than a general causal explanation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.