On the evening of October 26, 2023, the 58th Zhi·Xinforum, jointly organized bythe College of Electronic and Information Engineering, Tongji University and Shanghai Research Institute for Intelligent Autonomous Systems, was held online. Prof. Christian Ebenbauer from the RWTH Aachen University, Germany was invited to give a talk on "Control Theory and Optimization Algorithm Synthesis", which was moderated by Prof. Hao Zhang.
Firstly, Prof. Christian Ebenbauer introduced the motivation to study control theory and optimization algorithm synthesis, and presented some optimization methods in control as well as AI systems, including neural networks, support vector machines, reinforcement learning, adaptive control, and optimal control. He then described the reasons for analyzing and synthesizing optimization algorithms from a control perspective and pointed out the connection between control and algorithm synthesis problems. Prof. Christian Ebenbauer then showed two cases where algorithms were analyzed and synthesized from a control perspective. The first one was a convex synthesis of accelerated gradient descent algorithms based on robust control techniques and gives applications to a variety of cases, including optimal convergence rates, structural utilization, optimization of feedback systems, delay gradient measurements, and robust performance. The second was a gradient approximation based on nonlinear and optimal control, with specific analytical methods and simulation results given. Finally, Prof. Christian Ebenbauer summarized the synthesis of control theory to optimization algorithms and gave an outlook on the future direction of more control techniques to synthesize optimization algorithms.
After the talk, Prof.Christian Ebenbauer exchanged views with teachers and students on issues related to the synthesis of optimization algorithms, and encouraged students to broaden their horizons and explore, discover, and solve new scientific problems based on his own experience. This talk further broadened the horizons of our teachers and students and enhanced their knowledge and understanding of control theory synthesis optimization algorithms.
Written by Zhang Hao
Photographed by Wang Yunjiao