Trust, Confidence and Explainability in Artificial Intelligence. (3-0) 3 Credit Hours

Catalog description

Prerequisite: CS 3343 or consent of instructor. Study of fundamental methods for understanding strengths and weaknesses of machine learning and AI algorithms, including convolutional networks, recurrent neural networks, transformers and perceivers. Topics for explainability include attribution methods for AI, such as those based on gradients, Hessians, path integrals and Shapley values. Topics in trust and confidence include Platt scaling, temperature scaling, Bayesian networks and more modern calibration approaches based on attributions. Notions such as adversarial attacks, out-of-distribution (OOD) detection, and GANs will be discussed in the context of AI robustness. All concepts will be illustrated using real-world examples from both archival and contemporary literature. This course has Differential Tuition. Course Fees: GA02 $90; IUCA $45.