Title:  Designs of Network Architectures and Optimization Algorithms Based on Neural Differential Equations

Abstract:  This presentation investigates the theoretical and practical integration of deep neural networks and neural differential equations to address fundamental challenges in model robustness, trainability, and generalization. By interpreting network depth as time in dynamic systems, the research leverages numerical analysis concepts, specifically zero stability, consistency, and Shannon’s sampling theorem, to guide the design of stable and efficient network architectures. Furthermore, it introduces novel optimization algorithms, such as gradient activation, loss landscape reshaping, and integral-based smoothing, which utilize dynamic system principles to handle ill-conditioned problems, escape saddle points, and converge to flat minima. Collectively, we provide a low-cost, high-efficiency framework for enhancing deep learning performance by bridging discrete network structures with continuous differential equation theory.

Biography: Long Jin is a Professor and Ph.D. Supervisor at Lanzhou University and a recipient of the National Young Talent Program. In 2023, he served as a Visiting Professor in the Department of Computer Science at City University of Hong Kong. He has led four projects funded by the National Natural Science Foundation of China, one project under the National Key R&D Program of China, and multiple provincial and ministerial-level projects, including Key and Distinguished Young Scholar grants from the Natural Science Foundation of Gansu Province. He has been consecutively recognized as a Highly Cited Chinese Researcher by Elsevier and ranked among the top 0.05% of scientists worldwide. He has received several prestigious awards, including the Excellent Doctoral Dissertation Award from the Chinese Association for Artificial Intelligence, the Wu Wenjun Artificial Intelligence Award for Outstanding Young Scientist, and the Second Prize of the Natural Science Award of Gansu Province. His students have received numerous honors, including Excellent Doctoral and Master’s Thesis Awards from national academic societies and provincial authorities as well as the Excellent Doctoral Student Program funded by the National Natural Science Foundation of China. He currently serves as Associate Editor for several SCI journals, including IEEE Transactions on Industrial Electronics, IEEE Transactions on Intelligent Vehicles, IEEE Transactions on Automation Science and Engineering, IEEE Transactions on Fuzzy Systems, IEEE/CAA Journal of Automatica Sinica, Neural Networks, and CAAI Transactions on Intelligence Technology, and has received multiple Outstanding Editor Awards from IEEE/CAA JAS, CAAI TRIT, and IJCAS. His research interests include computational intelligence and its applications.