Reproduce “Recurrent In-Context Learning for 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
Tejas Vaidhya, Gaurav Sahu, Irina Rish
Tab-Loop classifies a new dataset by using its labelled rows as context for predicting unlabelled rows. It reuses four transformer blocks instead of applying twelve distinct blocks, with other components fixed. At equal training compute and twelve block applications, it reaches 84.10% TALENT mean accuracy, versus 83.92% for the baseline. Overlapping TabArena intervals bound the result to equivalence, not superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.