Research map / Label-efficient and robust learning / OOD and anomaly detection
ReTabAD: A Benchmark for Restoring Semantic Context in Tabular Anomaly Detection
Sanghyu Yoon, Dongmin Kim, Suhee Yoon, Ye Seul Sim, Seungdong Yoa, Hye-Seung Cho, Soonyoung Lee, Hankook Lee, Woohyung Lim
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 |
|---|---|---|---|
| add feature descriptions and domain knowledge to prompt (ablation) | Gemini-2.5-pro zero-shot · ReTabAD (20 datasets) | average AUROC: 0.691 -> 0.847 | Improved |
| add feature descriptions and domain knowledge to prompt (ablation) | GPT-4o-mini · ReTabAD (20 datasets) | average AUROC: 0.692 -> 0.726 | Improved |
| add feature descriptions and domain knowledge to prompt (ablation) | Qwen3-235B · ReTabAD (20 datasets) | average AUROC: 0.665 -> 0.747 | Improved |
| add metadata to prompt (ablation) | Gemini-2.5-pro · ReTabAD glioma | AUROC: 0.586 -> 0.895 | Improved |
| add metadata to prompt (ablation) | Gemini-2.5-pro · ReTabAD covertype | AUROC: 0.861 -> 0.961 | Improved |
| add metadata to prompt (ablation) | Gemini-2.5-pro · ReTabAD backdoor and automobile | AUROC: automobile 0.806 -> 0.767; backdoor 0.804 -> 0.827 | Mixed |
| add feature description to normal statistics (Type A -> B) (ablation) | zero-shot LLMs · ReTabAD (20 datasets) | win rate | Improved |
| use domain knowledge plus feature description without normal statistics (Type C) (ablation) | zero-shot LLMs · ReTabAD (20 datasets) | win rate | Mixed |
| add normal statistics to feature and domain context (Type C -> D) (ablation) | zero-shot LLMs · ReTabAD (20 datasets) | win rate: Gemini-2.5-pro win rate 0.15 -> 0.80 | Improved |
| add metadata to prompt (ablation) | Gemini-2.5-pro · ReTabAD glioma | F1@1 vs SHAP attribution: 0.009 -> 0.551 | Improved |
| provide reasoning examples of detected anomalies in the prompt (up to 5) | Gemini-2.5-pro · ReTabAD cirrhosis | AUROC | Improved |
Closest papers by meaning
- Beyond Individual Input for Deep Anomaly Detection on Tabular Data (ICML 2024 · similarity 0.87)
- ConTextTab: A Semantics-Aware Tabular In-Context Learner (NeurIPS 2025 · similarity 0.86)
- RADAR: Benchmarking Language Models on Imperfect Tabular Data (NeurIPS 2025 · similarity 0.86)
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · similarity 0.86)
- Enhancing Tabular Learners with Context-Aware Semantic Embeddings (NeurIPS 2026 · similarity 0.85)
- SetAD: Semi-Supervised Anomaly Learning in Contextual Sets (NeurIPS 2026 · similarity 0.85)
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.
