Reproduce “Two Is Enough: Binary Classification Priors Suffice for Multi-Class In-Context Tabular Prediction” 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

Zhixiong Cheng, Shiyuan Han, Jingjing Suo, Zhengli Wang

The authors evaluate TabICLv2-Custom-Binary-Prior, pretrained on synthetic binary classification tasks, on binary and multiclass datasets without updating its weights at inference. The paper proposes that labels in a table’s context determine prediction classes, but does not verify this mechanism. It reports 1590.1 Elo versus 1568.0 for the default; overlapping intervals and differing training recipes limit causal conclusions.

Summary by alphaXiv.

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