题目：A Genetic Algorithm for Optimal Design of Neural Networks

报告人：Prof. Jianqiao Sun（加州大学美熹德分校）
时间：2019年10月11日(星期五) 上午9:30
地点：明故宫校区A18-529
主办单位：航空学院、机械结构力学及控制国家重点实验室

报告内容简介：
 We are now in the age of data science when machine learning and artificial intelligence are beginning to reshape the traditional engineering research.  Examples of applications of machine learning, for example, are plenty already and are still coming out in large numbers.  In the machine learning, there is on theorem known as University Approximation Theorem which states that with a sufficient number of layers and neurons, any data set can be well modeled with a desired accuracy. Then, for a given data set, what is the smallest and best performing model for either regression and prediction or classification?  The question leads to a multi-objective optimization problem for designing optimal neural network architecture.

This talk introduces a genetic algorithm for optimal design of neural networks for modeling data sets.  We develop a special way to create genetic coding of networks, and algorithms to search for the optimal network design with minimum prediction error and minimum computational effort for a given data set.  Several examples are reported including the remaining useful life prediction of aircraft engine, and classification of images.  The advantage of having a light-weight network is its readiness for implementation on microprocessors.  This will lay a foundation for real-time applications of machine learning and artificial intelligence.

报告人简介：

Dr. Jianqiao Sun earned a BS degree in Solid Mechanics from Huazhong University of Science and Technology in Wuhan, China in 1982, a MS and a PhD in Mechanical Engineering from University of California at Berkeley in 1984 and 1988.  He worked for Lord Corporation at their Corporate R&D Center in Cary, North Carolina.  In 1994, Dr. Sun joined the faculty in the department of Mechanical Engineering at the University of Delaware as an Assistant Professor, was promoted to Associate Professor in 1998 and to Professor in 2003. He joined University of California at Merced in 2007, and is currently a professor and chair of the Department of Mechanical Engineering in School of Engineering.  He was elected to be a Qian-Ren-Ji-Hua scholar in the fourth cycle with Tianjin University.  Besides many other editorial experiences, he is the Editor-in-Chief of International Journal of Dynamics and Control published by Springer.

His research interests include stochastic non-linear dynamics and control, cell mapping methods, multi-objective optimization, high-density piezoelectric energy harvesting from highway traffic and data-driven energy management of office building HVAC systems.