The University of Wisconsin-Eau Claire invited Dr. Xin Zhang from the Department of Computer Science to present his research on computer vision in November. The virtual presentation underscored the critical importance of robustness in Convolutional Neural Networks (CNNs).
The presentation introduced the audience to a suite of quantitative tools for analyzing model robustness. These tools help identify cases where CNNs fail to generalize properly. For example, the talk identified challenges such as label missing and feature shifting and proposed effective solutions to fortify the robustness of CNNs against these issues.
Dr. Zhang’s talk emphasized that developing more resilient CNN models will be critical for deploying this technology safely.
