This research leverages Large Language Models (LLMs) to classify multivariate time series data when training examples are limited, addressing a critical challenge in industrial settings.

• Introduces LLMFew, a specialized framework that enhances few-shot classification performance
• Utilizes pre-trained LLM knowledge to compensate for limited training data
• Employs optimized encoders and fine-tuning approaches for time series data
• Demonstrates effectiveness across industrial and medical applications

Engineering Impact: This approach provides a practical solution for engineering teams faced with the common challenge of insufficient training data in industrial monitoring, predictive maintenance, and anomaly detection applications.

Large Language Models are Few-shot Multivariate Time Series Classifiers