Reproduce “A Transformer-Derived Iterative Preconditioner” 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
Patrick Lutz, Themistoklis Haris, Aditya Gangrade, Venkatesh Saligrama
RAPID preconditions SPD systems for CG through randomized coordinate mixing and sparse row corrections, without an inverse or factorization. The authors extracted this operation from a solution-predicting transformer and tested model-free RAPID on 31 SuiteSparse matrices. After 500 steps, mean condition number fell from 94,992.34 to 4,600.00, and mean CG iterations from 266.45 to 40.65; these averages do not establish runtime gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.