【摘要】We address the high dimensional data clustering issue with a simple neural network for effective performance of pattern recognition in lower dimensional subspaces and submanifolds. We suggest adaptive delay as an intermediate memory trace between the single neuron’s short term memory (activation) and the inter-neuron’s long term memory (synaptic weight), and we show how this adaption allows the data-driven neural network to adjust the learning rates in different dimensions of the input signals to dynamically select relevant features of emerging projective clusters. Experiments and applications will be given to illustrate the superiority of the obtained Projective Adaptive Resonance Theory (PART).

【报告人简介】Dr. Jianhong Wu is a University Distinguished Research Professor and a Canada Research Chair in Industrial and Applied Mathematics, York University, Canada. His research interest includes nonlinear dynamical systems, mathematical ecology and epidemiology, neural networks and complex data clustering. He has co-authored over 360 journal articles and 8 research books, in addition to a number of edited volumes and policy consultation reports.  He has served on the editorial boards of many international journals and is the co-editor for Infectious Disease Modelling. He is on the advisory/scientific boards of major organizations such as AARMS, BIRS, Fields Institute, NIMBioS.  Awards he received include the Queen's Diamond Jubilee medal from the government of Canada, the Canadian Applied and Industrial Mathematics Research Prize, the Cheung Kong Visiting Professorship, the Paul Erdos Visiting Professorship, and the Alexander von Humboldt Fellowship. He became a Fellow of the Fields Institute in 2011, and was the President of the Canadian Applied and Industrial Mathematical Society, 2011-2013. He received his PhD degree from Hunan University in 1987, and he was awarded the Doctor Honoris Causa by the University of Szeged in 2016.