Reproduce “ElTab: Elastically Looped Tabular Transformers” 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
Mieszko Komisarczyk, Saurabh Mathur, Maurice Kraus, Sriraam Natarajan, Kristian Kersting
ElTab lets tabular in-context classifiers adjust inference compute by choosing how many times a shared three-block stack is applied. During training, it learns predictions at shorter and full depths through distillation. Across 30 OpenML classification datasets with at most 2,000 rows, 100 features, and 10 classes, three loops score 0.8863 mean AUC-ROC versus 0.8890 for TabPFN, using 25% fewer inference FLOPs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.