Research map / Label-efficient and robust learning / OOD and anomaly detection
Beyond the Known: An Unknown-Aware Large Language Model for Open-Set Text Classification
Xi Chen, Chuan Qin, Ziqi Wang, Shasha Hu, Chao Wang, Hengshu Zhu, Hui Xiong
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 open-set (subset-conditioned K+1) fine-tuning instead of closed-set SFT (ablation) | LLaMA3.1-8B SFT · CLINC (25% known classes) | F1: 48.40 -> 75.50 | Improved |
| add contrastive learning (ablation) | Open-set SFT · CLINC (25% known classes) | K-F1: 75.24 -> 77.32 | Improved |
| add orthogonality constraint loss (ablation) | Open-set SFT + contrastive · CLINC (25% known classes) | K-F1: 77.32 -> 78.73 | Improved |
| add OOD parameter calibration (ablation) | Open-set SFT + contrastive + orthogonal · CLINC (25% known classes) | K-F1: 78.73 -> 80.09 | Improved |
| add analogy-augmented self-reflection (ablation) | Open-set SFT + contrastive + orthogonal + calibration · CLINC (25% known classes) | K-F1: 80.09 -> 83.01 | Improved |
| stack modules on closed-set SFT (contrastive, orthogonal, analogy) (ablation) | Closed-set SFT · CLINC (25% known classes) | F1: 48.40 -> 43.81 | Hurt |
| add contrastive learning to post-hoc LLM-OOD (ablation) | LLM-OOD · CLINC (25% known classes) | F1: 66.68 -> 30.21 | Hurt |
| replace analogy prompting of an untuned LLM with UnLLM fine-tuning | LLaMA3.1-8B · CLINC (25% known classes) | K-F1: 61.26 -> 83.90 | Improved |
| replace LLM-OOD with UnLLM | LLM-OOD · CLINC (50% known, representation separation) | Intra-Inter Ratio: 0.94 -> 21.30 | Improved |
| replace NPO with UnLLM | NPO · THUCNews (25% known classes) | K-F1: 61.06 -> 83.63 | Improved |
| increase lambda_cl to high values (ablation) | UnLLM · BANKING | K-F1 | Hurt |
Closest papers by meaning
- Idiosyncrasies in Large Language Models (ICML 2025 · similarity 0.87)
- Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection (NeurIPS 2025 · similarity 0.86)
- Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · similarity 0.86)
- Beyond the Seen: Bounded Distribution Estimation for Open-Vocabulary Learning (NeurIPS 2025 · similarity 0.85)
- 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)
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.
