报告题目: Bayesian Supervised Deep Deconvolutional Learning
报告人:Prof. Lawrence Carin,IEEE Fellow, Duke Vice Provost（美国杜克大学）
报告时间：2015年3月14日（周六） 早上09:00
报告地点:新科技楼1602会议室
Bio:
Lawrence Carin earned the BS, MS and PhD degrees from the University of Maryland, College Park, in 1985, 1986 and 1989, respectively. Since 1995 he has been with Duke University, where he is a professor of Electrical & Computer Engineering, with a research focus on machine learning. From 2011-2014 he was the Chair of the Duke ECE department, and now he serves as the Duke Vice Provost for Research. He held the William H. Younger Distinguished Professorship from 2004-2014, and he is an IEEE Fellow.
Abstract:
We develop a supervised model for deep (multilayered) deconvolutional learning, based on a Bayesian framework. The Bayesian support vector machine is linked to the top-layer features, yielding efficient bottom-up (pretraining) and top-down (refinement) max-margin discriminative learning. After learning the deep convolutional dictionary and classifier, testing is implemented via deconvolutional inference. To accelerate this inference, we project the top-layer dictionary elements to the data level, so only a single layer of deconvolution is required during testing. Classification results on the MNIST, Caltech 101, and Caltech 256 datasets demonstrate performance that exceeds or is very close to the previous state of the art.
主办单位：雷达信号处理重点实验室 西电国际合作与交流处
