Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·10d ago·5 min read
Have an LLM Write Your Anomaly Detector: Autonomous Discovery of Compact, Interpretable Detectors for Time Series
TL;DR
A large language model (LLM) was used to autonomously create two compact anomaly detectors for time series data, achieving superior performance without requiring GPU resources. These detectors excel in predictive accuracy and interpretability, outperforming traditional methods on the TSB-AD benchmark.
✦ Why It Matters
Consider using LLMs to automate the development of anomaly detection systems in your projects.
Key Takeaways
- Explore LLMs for automating anomaly detection development.
- Utilize the compact detectors for efficient time series analysis.
- Leverage local spectral features for improved interpretability.
- Adopt the covariance-aware distance metric for better accuracy.
How It Works
The LLM operates in an autonomous research loop, where it edits a short NumPy program to optimize the anomaly detection process. It evaluates the performance of each iteration based on a leakage-free objective, retaining the best-performing detector.
The resulting detectors analyze time series data by focusing on local spectral features and comparing them to the training data distribution using a covariance-aware distance metric.