您当前的位置:首页 >科学研究 > 学术交流

学术报告二十四: Learning-based PDEs: A Framework for Vision Model D

来源:数学与统计学院     作者:     时间:2017/9/20 17:00:35  0次

数学与统计学院学术报告[2017] 024

(高水平大学建设系列报告094)

讲座题目: Learning-based PDEs: A Framework for Vision Model Design and Connections to Deep Learning

讲座人:     刘日升 副教授  (大连理工大学

讲座时间:2017/9/24  上午9:00-10:30

讲座地点:科技楼515-4                                

报告内容:In vision society, PDEs have been widely used for image-level applications in the past decades. But due to the extremely high complexity and limited performance, conventional PDEs are rarely be considered for modern vision tasks. In this talk, we will introduce a new perspective, named Learning-based PDEs (LPDEs), to incorporate learning strategies to adaptively design PDEs for different kinds of vision tasks. Within this framework, we can use a single PDE system to address a variety of vision tasks. The discriminative and temporal information can also be investigated by our learned PDEs. Finally, we build underling connections between LPDE and deep learning models (e.g., CNN and ResNet) and provide insights and new models to address more challenging vision problems.

报告人简历

刘日升,大连理工大学副教授,博士生导师。近年来在计算机视觉与多媒体技术领域重要学术期刊(T-PAMIT-NNLST-MMMachine LearningPattern RecognitionNeural Networks等)和会议(CVPRNIPSAAAIECCVACM MMACCV等)发表论文六十余篇。相关工作目前被IJCVT-PAMIT-IP等期刊和IJCAIAAAICVPRICCVECCV等会议论文引用超过1000次,H-index13,最高单篇引用434次(Google Scholar)。担任IET Image ProcessingCCF推荐C类期刊)编委(Associate Editor),多次担任CVPRICCVECCVNIPSIJCAIAAAIACCVBMVCICIP等会议PC成员或审稿人以及IJCVT-PAMIT-IPT-NNLST-KDET-CSVTT-PDS等期刊审稿人。中国计算机学会多媒体专委会委员、中国计算机学会YOCSEF大连委员、中国图像图形学会机器视觉专委会委员,IEEE会员、ACM会员,CCF会员。

分享到: 0

数学与统计学院研究生工作奖励措施

2016年深圳大学数学与统计学院暨暑期夏令营招生指南