CLaSP introduces a novel approach for retrieving time-series signals using natural language descriptions, eliminating the need for sketch-based inputs or domain-specific dictionaries.

• Enables data scientists to search for complex time-series patterns using everyday language
• Leverages contrastive learning to align natural language descriptions with corresponding signal characteristics
• Demonstrates superior performance across multiple domains including industrial diagnostics
• Provides a scalable, domain-agnostic solution for signal retrieval tasks

This research revolutionizes how engineers interact with sensor data, offering practical applications in industrial monitoring, anomaly detection, and diagnostic systems where finding specific signal patterns is crucial.

CLaSP: Learning Concepts for Time-Series Signals from Natural Language Supervision