Speaker
Description
There is a 30-year history and leadership in verification, validation, and uncertainty quantification (VVUQ) established by the U.S. Department of Energy National Nuclear Security Administration (NNSA) Labs to evaluate the credibility of computational models used in high consequence applications. Today, the U.S. National Institute of Standards and Technology has established the building blocks for trustworthy artificial intelligence (AI) in response to the advancements in machine learning (ML). In this presentation we describe ongoing work to certify the trustworthiness of ML models by establishing the credibility evidence for an epidemiology example that leverages hybrid models, that include neural network model-form error corrections, as a novel diagnostic to elucidate global versus local trends in complex dynamical systems. We will introduce the variability in trustworthiness evidence and credibility in ML models is based on their datasets, inference goals, and training algorithms. As new emerging AI technology continues to revolutionize advancements in scientific discovery and
computational modeling, we articulate the importance in establishing a taxonomy of computational models. Prioritizing a taxonomy of AI algorithms, we establish the principles of AI trustworthiness for a class of algorithms that are well suited to the application context and provide a framework for establishing a certification process.
