StatVal: Statistical Validation for Computer Vision Models

The amount of data being collected by sensors and cameras has allowed for an explosion in the use of Artificial Intelligence and Machine Learning (AI/ML) to analyze and synthesize useful knowledge from streams of image data. These systems often act as black boxes to observers with little insight into how they function and little knowledge of how trustworthy their predictions and identification. Current approaches to the problem of image verification are typically done ad hoc and lack the systemic validation on many different types of image data. This poses a large problem when methods that work on one type of image data are applied to novel data types and scenes where they often break down. To address these problems, R-DEX Systems, Inc. (R-DEX) proposes StatVal, a groundbreaking technology built on previous R-DEX efforts to enable vision model validation and robustness. StatVal will be a single unified platform that will aid in hardening and verifying autonomous systems by: 1) identifying in what scenarios machine learning models are likely to stop functioning 2) creating machine learning models that are robust to natural and adversarial disturbances, and 3) building trust by understanding where the models are the most uncertain. In Phase I, R-DEX will demonstrate the scientific, technical, and commercial feasibility of StatVal and develop Phase II plans.