CIS 482: Trustworthy Artificial Intelligence

Course Description

This course introduces the broad and evolving notion of trustworthy artificial intelligence (AI). It covers three broad areas of trustworthiness in AI: robustness, transparency, and accountability. For robustness, the course introduces the AI threat landscape focusing on training data poisoning, model evasion, privacy-sensitive data inference, model stealing/extraction, and threats to safe deployment of AI. For transparency, the course covers frameworks used to interpret/explain AI model’s decisions. For accountability, the course discusses methods and tools for reducing bias and ethical pitfalls when AI models are deployed in high-stakes application domains. The course also discusses the dynamics among the three broad AI trustworthiness desirables. The course adopts a predominantly project-based setting to enhance hands-on experience. (W).