This technology is a methodological framework that utilizes actor-critic and quantile regression approaches to enhance anomaly detection.
Anomaly detection is a crucial task in various fields, including cybersecurity and quality control. Current methods of anomaly detection often rely on large amounts of labeled data and predefined thresholds, which are difficult to obtain and fine-tune. This creates challenges when anomalies are rare or when threshold selection leads to either missed detections or false alarms.
This technology uses an actor-critic-based approach and quantile regression to improve anomaly detection. With this method, the detection decision is chosen by the actor’s policy algorithm. Quantile regression helps to better estimate tail probabilities, which factor into the alarm decision-making process, allowing the system to more accurately select the threshold and correct outputs.
Patent Pending
IR CU24165
Licensing Contact: Greg Maskel
