20180801-23 李伏欣:Some Understandings and New Designs of Recurrent and

报告嘉宾：李伏欣（Corvallis, OR USA）

报告时间：2018年08月01日（星期三）上午10:00（北京时间）

报告题目：Some Understandings and New Designs of Recurrent and Convolutional Network Structures

主持人：杨猛（中山大学）

报告人简介：

Fuxin Li is currently an assistant professor in the School of Electrical Engineering and Computer Science at Oregon State University. Before that, he has held research positions in University of Bonn and Georgia Institute of Technology. He had obtained a Ph.D. degree in the Institute of Automation, Chinese Academy of Sciences in 2009. He (co-)won the PASCAL VOC semantic segmentation challenges from 2009-2012, and led a team to the 4th place finish in the DAVIS Video Segmentation challenge 2017. He has published more than 40 papers in computer vision, machine learning and natural language processing. His main research interests are deep learning, video object segmentation, multi-target tracking, structural prediction and human understanding of deep learning.

个人主页：

http://web.engr.oregonstate.edu/~lif/

报告摘要：

In this talk, I will talk about some of our recent work in understanding and reforming the well-known LSTM and CNN architectures. In the first part, I will talk about our experience utilizing LSTM in multi-target tracking and show some intuitions about why the current LSTM may be insufficient for long-term multi-object tracking. A novel bilinear LSTM model suitable for multi-target tracking problems will be proposed, motivated by the classic recursive least squares formulation. Results on the MOT 2016 and MOT 2017 challenges will be shown that significantly outperform traditional LSTMs in terms of identity switches. In the second part, I will talk about several of our recent work in understanding CNN from different angles, including a Gaussian complexity theory for understanding the generalization capabilities of the CNN and re-designing it for specific non-natural image problem domains, the explanation neural network (XNN) for generating human-understandable visual explanations of deep network decisions, and the PointConv approach we recently developed for implementing CNN on irregular point cloud data.

参考文献：

[1] Chanho Kim, Fuxin Li, James M. Rehg. Multiple Hypothesis Tracking Revisited. ICCV 2015.

[2] Chanho Kim, Fuxin Li, James M. Rehg. Multi-object Tracking with Neural Gating using bilinear LSTMs. ECCV 2018.

[3] Xingyi Li, Fuxin Li, Xiaoli Fern, Raviv Raich. Filter Shaping for Convolutional Networks. ICLR 2017.

[4] Zhongang Qi, Saeed Khorram, Fuxin Li. Embedding Neural Networks into Visual Explanations. arXiv:1709.05360 (preliminary version published on NIPS 2017 workshop on Interpreting, Explaining and Visualizing Deep Learning - Now what?)

[5] Wenxuan Wu, Zhongang Qi, Fuxin Li. PointConv: Deep Convolutional Networks on 3D Point Clouds. To appear on arXiv by 7/31.

18-23期VALSE在线学术报告参与方式：

长按或扫描下方二维码，关注”VALSE“微信公众号（valse_wechat），后台回复”23期“，获取直播地址。

特别鸣谢本次Webinar主要组织者：

VOOC责任委员：杨猛（中山大学）

VODB协调理事：王瑞平（中科院计算所）

活动参与方式：

1、VALSE Webinar活动依托在线直播平台进行，活动时讲者会上传PPT或共享屏幕，听众可以看到Slides，听到讲者的语音，并通过聊天功能与讲者交互；

2、为参加活动，请关注VALSE微信公众号：valse_wechat 或加入VALSE QQ群（目前A、B、C、D、E、F、G群已满，除讲者等嘉宾外，只能申请加入VALSE H群，群号：701662399）；

*注：申请加入VALSE QQ群时需验证姓名、单位和身份，缺一不可。入群后，请实名，姓名身份单位。身份：学校及科研单位人员T；企业研发I；博士D；硕士M。

3、在活动开始前5分钟左右，讲者会开启直播，听众点击直播链接即可参加活动，支持安装Windows系统的电脑、MAC电脑、手机等设备；

4、活动过程中，请不要说无关话语，以免影响活动正常进行；

5、活动过程中，如出现听不到或看不到视频等问题，建议退出再重新进入，一般都能解决问题；

6、建议务必在速度较快的网络上参加活动，优先采用有线网络连接；

7、VALSE微信公众号会在每周一推送上一周Webinar报告的总结及视频（经讲者允许后），每周四发布下一周Webinar报告的通知及直播链接。

[slides]