Research map / Large language models / Language, safety and interpretability
LLM as an Algorithmist: Enhancing Anomaly Detectors via Programmatic Synthesis
Hangting Ye, Jinmeng Li, He Zhao, Mingchen Zhuge, Dandan Guo, Yi Chang, Hongyuan Zha
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 LLM-DAS synthetic hard anomalies | OCSVM · 36 tabular AD datasets | AUC-PR: +0.0723 AUC-PR (84.49%) | Improved |
| add LLM-DAS synthetic hard anomalies | PCA · 36 tabular AD datasets | AUC-PR: .5975 -> improvement .0402 (21.50%), win 30 / 36 | Improved |
| add LLM-DAS synthetic hard anomalies | IForest · 36 tabular AD datasets | AUC-ROC: .7555 -> improvement .0385 (10.90%), win 28 / 36 | Improved |
| add LLM-DAS synthetic hard anomalies | ECOD · 36 tabular AD datasets | AUC-PR: .5376 -> improvement .0512 (23.05%), win 28 / 36 | Improved |
| add LLM-DAS synthetic hard anomalies | DRL · 36 tabular AD datasets | AUC-ROC: .8509 -> improvement .0272 (3.79%), win 21 / 36 | Improved |
| use OCSVM-specific instead of PCA-specific synthesis code (ablation) | OCSVM · Cross-detector synthesis on OCSVM (7 datasets) | AUC-PR average change: -0.0213 -> +0.0295 | Improved |
| use OCSVM-specific instead of IForest-specific synthesis code (ablation) | OCSVM · Cross-detector synthesis on OCSVM (7 datasets) | AUC-PR average change: +0.0076 -> +0.0295 | Improved |
| use mismatched PCA code on OCSVM (ablation) | OCSVM · Vowels | AUC-PR change: -0.1899 | Hurt |
| replace detector-aware prompt with generic prompt, no-score-access prompt, or random seed selection (ablation) | LLM-DAS · 36 tabular AD datasets | AUC-PR | Improved |
| use random outlier synthesis instead of LLM-DAS | DRL · 36 tabular AD datasets | AUC-PR | Hurt |
Closest papers by meaning
- Beyond Individual Input for Deep Anomaly Detection on Tabular Data (ICML 2024 · similarity 0.86)
- AD-FM: Multimodal LLMs for Anomaly Detection via Multi-Stage Reasoning and Fine-Grained Reward Optimization (AAAI 2026 · similarity 0.86)
- Local Sparsity Enables Unsupervised LLM Safety Detection (NeurIPS 2026 · similarity 0.86)
- Tab-Shapley: Identifying Top-k Tabular Data Quality Insights (AAAI 2025 · similarity 0.86)
- Combating Phone Scams with LLM-based Detection: Where Do We Stand? (Student Abstract) (AAAI 2025 · similarity 0.85)
- DABL: Detecting Semantic Anomalies in Business Processes Using Large Language Models (AAAI 2025 · 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.
