"Immunization and Test-Time Augmentation for Pre-Trained Computer Vision Models" - Raymond Yeh

Originally presented on: Monday, December 9th, 2024 at 11:30pm CT, TTIC, 6045 S. Kenwood Avenue, 5th Floor, Room 530
Title: "Immunization and Test-Time Augmentation for Pre-Trained Computer Vision Models"
Speaker: Raymond Yeh, Purdue University
Abstract: With the advancement of open-source pre-trained models, computer vision systems are built largely by starting from these models followed by a fine-tuning/adaptation stage. In this context, we explore two important aspects of pre-trained models: (a) the mitigation of the potential misuse of pre-trained models, and (b) methods to improve the performance of a pre-trained model at test time. In this talk, I will present two recent works from our lab addressing these two topics, namely, "IMMA: Immunizing Text-to-Image Models Against Malicious Adaptation" and "Deep Networks with Subsampling Layers Unintentionally Discard Useful Activations at Test Time."
Short Bio: Raymond A. Yeh is an Assistant Professor in Computer Science at Purdue University. His research focuses on the intersection of machine learning and computer vision. He develops algorithms to learn effective computer vision models, with a recent focus on immunizing pre-trained vision models against fine-tuning and developing vision models with provable equivariance guarantees. Before joining Purdue, Raymond spent a year as a research assistant professor at the Toyota Technological Institute at Chicago (TTIC). He holds a PhD and an M.S. in ECE from the University of Illinois at Urbana-Champaign (UIUC). Raymond is a recipient of the Google PhD Fellowship, the Mavis Future Faculty Fellowship, and the Henry Ford II Scholar Award.
Timestamps:
00:00
00:05 Intro
00:53 Lecture
52:15 Q&A
#artificialintelligence #ai #machinelearning #algorithm #computervision #robotics #research