This research explores how multimodal large language models can help automate and improve time series anomaly detection systems by identifying false alarms and enhancing human oversight processes.

• LLMs effectively integrate visual data to identify false alarms in time series data
• Applications span healthcare, finance, manufacturing, and security domains
• Reduces need for constant human oversight while maintaining high accuracy
• Creates a human-AI partnership model for more efficient anomaly detection

In healthcare settings, this approach could dramatically improve patient monitoring systems by filtering out false alarms that lead to alert fatigue, allowing medical professionals to focus on genuine anomalies requiring intervention.

Refining Time Series Anomaly Detectors using Large Language Models