时间：10月31日（周四）11：00-12：00

地点：勤园21号楼306学术报告厅

主讲人：邓世文，教授，哈尔滨师范大学数学91制片 ，2012年毕业于哈尔滨工业大学计算91制片 人工智能与信息处理专业，主要研究方向：优化方法与智能信息处理、模式识别、机器学习、信号处理等，具体涉及以下应用研究内容：声学事件检测、声音场景识别、心音信号分析与识别、遥感图像处理与识别、信号增强等。在IEEE T. SIGNAL. PROCES.、IEEE SIGNAL PROC. LET.、MECH. SYST. SIGNAL PR.、DIGIT SIGNAL PROCESS.等期刊及本领域国际学术会议ICASSP、Interspeech、ICPR等发表学术论文19篇。

内容简介：The manifold assumption is wildly used in maching learning, based on which many mainifold-inspired methods are described, explicitly or Implicitly.

In this report, we provide some results of the autoencoder (AE) in deep learning and our current works as follows:

1) explicit geometric interpretation in contractive AE, denosing AE, and winner-take-all AE;

2) Implicit manifold assumption in sparse coding

3) Fast approximation for primal splitting algorithm with the manifold assumption.