数学与系统科学研究院

计算数学所学术报告

报告人：        Wei Wu

Dlian University of Technology, China

报告题目 Neural Network Computing

Abstract: Prof. Wei Wu has done a profound study on the convergence of online g radient method for neural network (OGM). Most existing convergence results for OG M are of probabilistic nature, saying that a weight learning iteration procedure will “very likely” converge under suitable conditions. Wu follows another appro ach to investigate OGM, and proves a series of convergence results of determinist ic nature, i.e., a weight learning iteration procedure will “definitely” conver ge under some special conditions. He has also considered the influence of the mom entum term (used to accelerate the convergence and improve the stability) and the penalty term (used to prevent the weights from getting too large) to the converg ence of OGM, and some applications of OGM in the prediction of stock market and t he recognition of mathematics formulas.

报告时间：2005年3月23日  下午4：00-5:00

报告地点：科技综合楼三层报告厅