Research map / Learning theory and algorithms / Prediction and decision losses
Algorithmic Recourse of In-Context Learning for Tabular Data
Wenshuo Dong, Jiaming Zhang, Shaopeng Fu, Hongbin Lin, Di Wang, Lijie Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ASR-ICL on ICL models instead of DiCE or FACE on trained models | TabPFN-2.5 · Australian Credit | average recourse cost: 8.96 (DiCE) and 6.18 (FACE) -> 3.83 | Improved |
| compare ASR-ICL with AR (cheaper but lower validity) | ASR-ICL · Australian Credit | validity: 0.82 (AR) -> 1.00 (TabPFN-2.5) | Improved |
| increase number of in-context instances (ablation) | ASR-ICL across ICL models · Australian Credit, COMPAS | validity and average cost | Improved |
| use adaptive subspace selection instead of full-space zeroth-order search (Full ZO) (ablation) | ASR-ICL · Australian Credit, Corporate Rating | query budget: roughly 2-3x fewer queries | Improved |
| use adaptive subspace selection instead of Full ZO (ablation) | ASR-ICL · Australian Credit, Corporate Rating | recourse cost | Improved |
| resample 32-shot context instances independently (ablation) | ASR-ICL · Australian Credit, COMPAS | validity and cost variance | No clear effect |
| vary trade-off parameter lambda and subspace size k moderately (ablation) | ASR-ICL · sensitivity analysis | recourse quality | No clear effect |
| use general LLMs instead of specialized tabular ICL models for multi-class recourse | ASR-ICL · Student Performance | validity: 1.00 (TabPFN-2.5) -> 0.42 (Qwen3-8B) | Hurt |
| use GPT-4o as ICL predictor | ASR-ICL · COMPAS and Diabetes | validity: 0.78 and 0.71 (GPT-4o) vs 1.00 (TabPFN-2.5) | Hurt |
Closest papers by meaning
- TabDPT: Scaling Tabular Foundation Models on Real Data (NeurIPS 2025 · similarity 0.90)
- TabICL: A Tabular Foundation Model for In-Context Learning on Large Data (ICML 2025 · similarity 0.90)
- ConTextTab: A Semantics-Aware Tabular In-Context Learner (NeurIPS 2025 · similarity 0.90)
- Mixture of In-Context Prompters for Tabular PFNs (ICLR 2025 · similarity 0.88)
- TabFlex: Scaling Tabular Learning to Millions with Linear Attention (ICML 2025 · similarity 0.87)
- Understanding the Surprising Generalization Properties of Tabular Foundation Models (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
