Research map / Label-efficient and robust learning / OOD and anomaly detection
Likelihood Ratios and Generative Classifiers for Unsupervised Out-of-Domain Detection in Task Oriented Dialog
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 generative-classifier likelihood instead of simple language-model likelihood (ablation) | OOD detection · ROSTD | F1: 81.38 ± 0.19 -> 86.25 ± 0.71 | Helped |
| Add a background-model likelihood ratio to generative-classifier likelihood (ablation) | L_gen OOD detection · ROSTD | F1: 86.25 ± 0.71 -> 91.35 ± 2.62 | Helped |
| Add a background-model likelihood ratio to generative-classifier likelihood (ablation) | L_gen OOD detection · SNIPS, 75% | F1: 67.31 ± 7.06 -> 85.47 ± 6.90 | Helped |
| Use square-root unigram noise rather than uniform noise for the background model (ablation) | L_simple likelihood-ratio OOD detection · ROSTD | F1: 85.25 ± 0.72 -> 87.42 ± 0.45 | Helped |
| Use square-root unigram noise rather than uniform noise for the background model (ablation) | L_gen likelihood-ratio OOD detection · ROSTD | F1: 89.60 ± 0.56 -> 91.17 ± 0.32 | Helped |
| Use unigram noise rather than uniform noise for the background model (ablation) | L_gen likelihood-ratio OOD detection · SNIPS, 75% | F1: 78.37 ± 6.60 -> 85.47 ± 6.90 | Helped |
| Increase MSP softmax temperature to 1e^3 (ablation) | MSP OOD detection · ROSTD | F1: 54.22 ± 4.01 -> 55.45 ± 4.19 | Helped |
| Use likelihood ratios rather than plain likelihood (ablation) | Likelihood-based OOD detection · ROSTD | F1: 81.38 ± 0.19 -> 87.42 ± 0.45 | Helped |
| Use the generative-classifier likelihood ratio rather than plain generative-classifier likelihood (ablation) | Likelihood-based OOD detection · ROSTD-COARSE | F1: 86.67 ± 0.34 -> 90.14 ± 0.39 | Helped |
Closest papers by meaning
- Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection (NeurIPS 2025 · similarity 0.86)
- Meta-Crafting: Improved Detection of Out-of-Distributed Texts via Crafting Metadata Space (Student Abstract) (AAAI 2024 · similarity 0.86)
- Training-free LLM-generated Text Detection by Mining Token Probability Sequences (ICLR 2025 · similarity 0.85)
- SeqGPT: An Out-of-the-Box Large Language Model for Open Domain Sequence Understanding (AAAI 2024 · similarity 0.85)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.85)
- Concept Matching with Agent for Out-of-Distribution Detection (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.