"Computer Vision Beyond Task Performance" – Raymond Yeh, TTIC Colloquium

“Computer Vision Beyond Task Performance” Raymond A. Yeh, Purdue University Originally recorded on April 6, 2026, at TTIC. In this talk, Raymond Yeh explores approaches to computer vision that move beyond benchmark accuracy, focusing on robustness, security, and consistency in learned models. He introduces model immunization as a defense against unauthorized fine-tuning and examines equivariance as a framework for ensuring predictable transformations in model outputs, advancing the design of more reliable vision systems. Timestamps: 00:00 Introduction 01:45 Talk begins 56:50 Q&A #ComputerVision #MachineLearning #AI #Equivariance #ModelSecurity #DeepLearning #Robustness #Research #TTIC #Colloquium