arXiv cs.AI· Xiaona Zhou, Muntasir Wahed, Tianjiao Yu, Constantin Brif, Ismini Lourentzou·· 4 小时前AI 评分34
小而可信：VisAnomReasoner 高效视觉-语言推理用于时间序列异常检测
Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection
AI 导读
研究者构建了 VisAnomBench 基准并基于其微调出参数高效的视觉-语言模型 VisAnomReasoner，用于时间序列异常检测。该模型在 VisAnomBench 上超过所有基线，precision 和 F1 分别至少提升 21.23 和 23.87 个百分点；在 TSB-AD-U 基准上 precision 和 F1 分别提升 9.57 和 13.39 个百分点，展现跨基准泛化能力。
正文 · 原文
Abstract:Recent advances in Vision--Language Models (VLMs) have achieved impressive performance across many tasks, yet prior studies report unsatisfactory performance when applying large language or multimodal models to finding abnormal patterns in sequential data. Public anomaly detection benchmarks typically provide interval annotations but not natural-language rationales, making it difficult to fine-tune VLMs to produce grounded, interpretable decisions. To address this gap, we construct VisAnomBench, a curated benchmark built from public time-series datasets and augmented with high-quality anomaly explanations selected from multiple large VLMs using fine-grained, task-specific rewards. Through fine-tuning on this benchmark, we develop VisAnomReasoner, a parameter-efficient VLM for time-series anomaly detection. Results on VisAnomBench show that VisAnomReasoner achieves more accurate anomaly localization and consistently outperforms all baselines, with improvements of at least 21.23 and 23.87 percentage points in precision and F1, respectively. Additional experiments on the TSB-AD-U benchmark demonstrate strong cross-benchmark generalization, with VisAnomReasoner improving precision and F1 by 9.57 and 13.39 percentage points, respectively.
| Comments: | Accepted at NeurIPS 2026. Updated experiments and appendix |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.30344 [cs.AI] |
| (or arXiv:2605.30344v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.30344 arXiv-issued DOI via DataCite |
Submission history
From: Xiaona Zhou [view email]
[v1]
Thu, 28 May 2026 17:59:50 UTC (2,579 KB)
[v2]
Thu, 5 Nov 2026 03:32:45 UTC (2,049 KB)
来源：arXiv cs.AI · arxiv.org