竞博电竞app
竞博电竞app
当前位置:首页  学术报告
竞博百年校庆竞博电竞系列学术报告之六十三 北京大学董彬副教授学术报告通知
发布人:蔡易  发布时间:2019-11-12   浏览次数:10

应竞博电竞张达治副教授邀请,北京大学北京国际数学研究中心董彬副教授将于近日来访我校并做一场关于图像处理反问题的数学和深度学习方法学术报告。以下是报告信息,欢迎各位师生参加。

时间:20191114日(星期四)14:00-16:00

地点:格物楼503报告厅

题目:Mathematical and Deep Learning Methods for Inverse Problems in Imaging

摘要:Inverse problems in imaging, including image denoising, deblurring, inpainting, computed tomography, etc., is one of the essential areas in imaging science. In image restoration, the wavelet-based approach and PDE-based approach have been particularly successful and widely adopted in both academia and industry. In recent years, deep learning methods start to dominant imaging, as well as many other fields where data is relatively abundant. While the model design of wavelet- and PDE-based approaches mostly relies on human wisdom (so-called handcraft modeling), the model design of deep learning approach primarily relies on data. There are increasing interests in the field in combing handcraft modeling and data-driven modeling so that we can enjoy the merits of both approaches.  

In this talk, I will start with a review of our previous works that established rigorous and generic connections between wavelet- and PDE-based approaches. This includes links of the wavelet-based optimization models to the total variation model and the Mumford-Shah model. Furthermore, links of wavelet frame shrinkage to nonlinear evolution PDEs are also established. Then, I will summarize our more recent work on bridging deep learning with differential equations and optimal control and demonstrate the advantage of such a bridge in modeling and algorithmic design for various tasks in image and shape analysis. In particular, I will show how numerical schemes of differential equations and optimization algorithms can lead to effective deep network architectures for medical imaging, low-level vision, and natural language processing.  

报告人概况:董彬,北京大学,北京国际数学研究中心长聘副教授、主任助理。2009年在美国加州大学洛杉矶分校数学系获得博士学位。博士毕业后曾在美国加州大学圣迭戈分校数学系任访问助理教授、20112014年在美国亚利桑那大学数学系任助理教授,2014年底入职北京大学。主要研究领域为应用调和分析、反问题计算、深度学习及其在图像和数据科学中的应用。在理论上,与合编辑一起将图像领域独立发展近30年的两个数学分支(PDE/变分方法和小波方法)建立深刻的联系,改变了领域内对这两类方法的认识,拓宽了应用范畴。应用上,以数学理论为引导思想,为来源于医学影像、计算机视觉、深度学习等领域中的重要问题提供行之有效的解决方案。在国际重要学术期刊和会议上发表论文60余篇,现任期刊《Inverse Problems and Imaging》编委。于2014年获得香港求是基金会的求是杰出青年学者奖,2015年入选中组部第十一批“千人计划”青年人才。


XML 地图 | Sitemap 地图