Domain Neutral Platform for Anomaly Detection in Autonomous Systems

The objective of this work is to transition the prototype anomaly detection system for ML/AI models,áCanary, into a standalone, easily integrable solution for ensuring deployed ML/AI models operate within the boundaries of their competency. The prototype solution,áCanary, does this by learning to distinguish data which can be reasonably processed by the coupled ML/AI model from data for which the ML/AI model is not suited and would likely produce erroneous or poorly calibrated results. Under the proposed effort, the prototype system is to be served with a web-based microservice to ease integration of the prototype with actively deployed systems and to provide a unified dashboard for monitoring model health. Additionally, Etegent seeks to study how the prototype system could be used to predict the performance characteristics of coupled ML/AI models.??????