일시: 2026. 8. 12. (수요일), 오후 3시

장소: 제4공학관 D504호

Speaker: Grigorios Chrysos, Ph.D. / Assistant Professor, University of Wisconsin-Madison

Homepage: https://grigoris.ece.wisc.edu

Title: Anatomy of a Hallucination: How Diffusion Solvers Shape Generation Errors

Abstract:
Diffusion models consistently produce stunningly realistic images, but they also frequently synthesize images that hallucinate. Building a reliable pipeline free of structural hallucinations forces us to ask key questions: can we prove that diffusion models will never hallucinate? If they necessarily hallucinate, does stochastic or deterministic sampling differ in their hallucination rate? In the second part of the talk, we flip the question: are hallucinations always detrimental?

Bio:
Dr. Chrysos is an Assistant Professor at the University of Wisconsin-Madison. His research focuses on trustworthy and reliable machine learning, specifically targeting the understanding and mitigation of hallucinations in generative models. Prior to joining UW-Madison, he was a Postdoctoral Researcher at EPFL (École Polytechnique Fédérale de Lausanne) working with Volkan Cevher. He earned his PhD from Imperial College London under the supervision of Stefanos Zafeiriou, during which he completed research internships at Facebook, NVIDIA Research, and Preferred Networks.