MIT researchers have developed a method using large language models to detect anomalies in complex systems without the need for training. The approach, called SigLLM, converts time-series data into text-based inputs for the language model to process. Two anomaly detection approaches, Prompter and Detector, were developed and showed promising results in initial tests.
The relationship between predictability and reconstructability, and how it can vary in opposite directions in complex systems. The work is based on information theory and was performed on various dynamics on random graphs, including continuous deterministic systems, and provides analytical calculations of the uncertainty coefficients for many different systems.
