All Times ET
Keywords: AI Assurance, Attack Localization, Deep Learning, Game Theory, Model-Agnostic
In this manuscript, we present Adversarial Logging Scoring Pipeline (ALSP): an Artificial Intelligence Assurance (AIA) set of algorithms. ALSP is model-agnostic, independent of the domain (i.e. healthcare, energy, banking), and provides scores for AIA goals including explainability, safety, and security. ALSP is scalable and tested with a use case (water distribution network) to illustrate it’s benefits. The pipeline (consists of three algorithms) optimizes AI models using a game theory approach; it also logs and scores the actions of the model to detect adversarial inputs.