Reproduce “When Repeated Measurements Hurt 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
Yixiang Liu, Ruizhe Deng, Bibhas Chakraborty, Qiong Zhang, Yan Shuo Tan
The authors study how tabular foundation models use repeated measurements at fixed input sites. They compare reading rows with one row per site across synthetic, semi-synthetic, and real tasks, and analyze protocol mismatch and TabPFN-v3 count resets. Harm is more frequent when within-site variation is low, but effects vary by model and setting; real-data targets are held-out site means.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.