Research map / Label-efficient and robust learning / OOD and anomaly detection
MISF: MLLM Guided Iterative Sample Filtering for Data Fault Detection
Guoying Chen, Ruizhuo Zhao, Zhewei Xu, Bo Yang, Kunlong Wang
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 curated clean subset to generated images for initialization (ablation) | MISF · RESISC45 symmetric 20% | FPR: 0.109 -> 0.071 | Improved |
| add generated images to curated clean subset for initialization (ablation) | MISF · RESISC45 symmetric 20% | FPR: 0.089 -> 0.071 | Improved |
| add generated images to curated subset at 40% noise (ablation) | MISF · RESISC45 symmetric 40% | FPR: 0.048 -> 0.040 | No clear effect |
| select by Gini impurity instead of confidence (ablation) | MISF · RESISC45 symmetric 20% | FPR: 0.039 -> 0.043 | Mixed |
| select by Gini impurity instead of confidence (ablation) | MISF · Oxford symmetric 20% | TPR: 0.994 -> 0.997 | Improved |
| use MISF instead of Cola | MISF · asymmetric noise 40% | FPR: 0.088 -> 0.048 | Improved |
| use MISF instead of MLLM zero-shot | MISF · symmetric noise 20% | FPR: 0.495 -> 0.069 | Improved |
| use MISF instead of spectral signature for poisons | MISF · RESISC45 poison 10% | FPR: 0.501 -> 0.036 | Improved |
| train on MISF-purified data vs noisy data | MISF · Oxford asymmetric 20% | accuracy: 0.8059 -> 0.9540 | Improved |
| increase Gini threshold (ablation) | MISF · RESISC45/Oxford | TPR/FPR | Mixed |
Closest papers by meaning
- Contamination Detection for VLMs Using Multi‑Modal Semantic Perturbations (ICLR 2026 · similarity 0.82)
- The Power of Iterative Filtering for Supervised Learning with (Heavy) Contamination (NeurIPS 2025 · similarity 0.81)
- Robust Conformal Outlier Detection under Contaminated Reference Data (ICML 2025 · similarity 0.81)
- Binary Decomposition: A Problem Transformation Perspective for Open-Set Semi-Supervised Learning (ICML 2024 · similarity 0.81)
- TDDBench: A Benchmark for Training data detection (ICLR 2025 · similarity 0.80)
- AD-FM: Multimodal LLMs for Anomaly Detection via Multi-Stage Reasoning and Fine-Grained Reward Optimization (AAAI 2026 · similarity 0.80)
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.
