AIFT Seminar by Prof. Dingxuan ZHOU

A seminar titled “The Role of Structures in Deep Neural Networks” took place on 3 Jan 2025. Prof. Dingxuan Zhou, Head of the School of Mathematics and Statistics at the University of Sydney, delved into the evolving landscape of neural networks structures. The classical mathematical theory, developed 35 years ago for fully-connected neural networks, falls short in addressing the intricacies of structured networks employed in modern deep learning applications like speech recognition, computer vision, and natural language processing. Prof. Zhou shared insights on the essential distinctions between fully-connected neural networks and structured ones, particularly focusing on deep convolutional neural networks. Prof. Zhou also discussed the generalization properties of deep learning algorithms induced by structured neural networks, offering invaluable insights into their diverse applications.