Reproduce “CCD: Context-Calibrated Open-Entity Out-of-Distribution Detection in In-Context 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
Zhangkai Wu, Xu Zhang, Zhihao Hao, Zhihong Cui, Yongmin Yoo, Xianxun Zhu, Kaize Shi
CCD detects unknown classes while accepting known-class rows from new entities in open-entity out-of-distribution detection. It calibrates confidence and three distance scores using known-class rows from held-out groups, then weights and combines the scores; the model remains frozen and no out-of-distribution data are used. Across 14 tables, CCD reaches mean AUROC 81.8%, versus 74.4% for kNN†, within the evaluated setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.