Reproduce “Beyond Row-Level Prediction: Evaluating and Learning Composition for Reusable Table Representations” 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

Ali Younes, Saeed Ghoorchian, Maximilian Schambach, Georgia Chalvatzaki, Johannes Höhne

The paper evaluates whole-table embeddings for matching partial table views and retrieval. It benchmarks overlap consistency, label-based separation, robustness, and throughput, then learns pooling heads over frozen ConTextTab states using overlap-guided training. On synthetic data, attention pooling raises D1 from 0.04 to 0.50, but fingerprints remain strongest in some settings, and the comparison does not isolate architecture from objective.

Summary by alphaXiv.

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