Abstract: Generative models can now create images so realistic they challenge human detection. In this talk on Fiday, April 10, at 11 a.m. in Building 4, Room 406, Dr. Ojha discusses his work on enabling these models to generate novel visual concepts that fall outside their original training such as generating a car with a dog’s furry texture. He will also share methods for detecting AI-generated fakes and discuss his current research interests concerning the fundamental nature of AI-generated data; what it is, and what it cannot be.

Utkarsh Ojha is an assistant professor at Bellini College of Artificial Intelligence, Cybersecurity and Computing in the University of South Florida. Dr. Ojha has served as a postdoctoral fellow at Carnegie Mellon University. He received his Ph.D. in computer science from the University of Wisconsin-Madison.

His research work has centered around generative models for images. He operates with the principle that finite mathematical models cannot simulate the visual or textual reality of the natural world. His academic goal is to show how and why this is the case even for the most advanced machine learning models.