TimeRLM: Recursive Language Models Enable Precise Anomaly Localization in Long-Context Time-Series

2026-08-04

AI Summary

This work addresses the challenging task of anomaly localization in long-context time series, where critical evidence is often embedded within long-range dependencies of high-frequency data. To tackle this, we propose TimeRLM—the first recurrent language model specifically designed for time series anomaly localization—integrating code-driven signal operations, vision-assisted comprehension, an external context querying mechanism, and reinforcement learning-based post-training. We also introduce AnomalyXL, a synthetic benchmark for systematic evaluation. Experimental results demonstrate that TimeRLM achieves an IoU of 0.682 and a classification-with-evidence accuracy of 0.745 on AnomalyXL-Localize, substantially outperforming existing time series language models, while maintaining or even improving performance on real-world ECG, sleep, and software monitoring datasets.

Abstract

Precise anomaly localization over long-context time series is a crucial task in monitoring applications across clinical care, industrial operations, financial services, and logistics, where brief evidence may hide inside long spans of high-frequency data. Time-Series Language Models (TSLMs) are able to ingest time series data and verbalize findings on anomalies in natural language; however, recent benchmarks report a decrease in retrieval performance at long contexts, mirroring failure modes in text, vision, and audio. In the text domain, Recursive Language Models (RLMs) can recover much of this lost performance by keeping context external to the large language model (LLM), allowing the model to query it through code. We present TimeRLM, an RLM formulation for time-series that sequentially manipulates the signal using code and vision capabilities. We further introduce AnomalyXL, a synthetic long-context anomaly localization benchmark with programmatically injected anomalies that require precise retrieval. We implement five different task categories and two variants: AnomalyXL-MCQ and AnomalyXL-Localize. TimeRLM outperforms every evaluated TSLM and single-pass baseline on four of the five AnomalyXL-Localize tasks, reaching 0.682 IoU on localization and 0.745 on classify-with-evidence, versus at most 0.329 and 0.072 across all baselines. We post-train TimeRLM using reinforcement learning. The resulting model further improves performance and requires approximately one-third as many agent interaction turns as its untrained base model to produce a final answer. On unseen real-world ECG, sleep and software observability recordings, the post-trained TimeRLM retains or improves performance, surpassing TSLMs despite being trained exclusively on synthetic data. Our findings suggest recursive interaction with time-series is an effective approach for long-horizon retrieval.

Problem

• anomaly localization
• long-context time series
• time-series monitoring
• precise retrieval
• high-frequency data

Innovation

• Recursive Language Models
• Time-Series Anomaly Localization
• Long-Context Retrieval
• Reinforcement Learning Post-Training
• Synthetic Benchmark