Reproduce “FTFM: Factorized Tabular Foundation Models” 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
Josif Grabocka
FTFM predicts labels for new table rows in one forward pass, without updating its weights on the target table. It retains cell representations while routing interactions through row and column summaries. On BeyondArena’s 98 classification datasets, it ranks first among 29 methods, with Elo 1280 and mean rank 7.9; this does not establish universal wins or isolate architecture’s effect. Measured prediction-time scaling is comparable to collapsing models, but absolute runtime comparisons use different hardware.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.