Reproduce “TILE: Bridging LLMs and Table Foundation Models for Tabular Learning with Strings” 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

Félix Lefebvre, David Holzmüller, Gioia Blayer, Myung Jun Kim, Marine Le Morvan, Gaël Varoquaux

TILE combines language-model representations with a tabular learner for tables whose strings carry predictive signal. It serializes each row with column names and values, then adds pooled language-model features alongside numerical and categorical features. On 65 regression datasets with 1,024 training rows, it scores 91.76 versus 77.44 for cell-wise encoding on a normalized R-squared scale multiplied by 100.

Summary by alphaXiv.

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