Some Understandings and New Designs of Convolutional and Recurrent networks

题  目:  Some Understandings and New Designs of Convolutional and Recurrent networks

主讲人:  Prof. Fuxin Li

时  间:  2019年08月29日(周四)上午10:00

地  点:  北京大学UU快三大楼106报告厅 


This talk will present some of our recent work in trying to understand the well-known convolutional and recurrent networks and explore some new designs. We start with explaining convolutional networks, where I will talk about our approaches in explaining deep network predictions to human. Drawbacks in traditional CNNs, such as rigidity of the convolutional filter, and overconfidence on outliers and adversarial examples lead us to explore new designs of the CNN. We will talk about PointConv, which efficiently implements CNN on irregularly spaced 3D point cloud

data, as well as HyperGAN, where we generate all the weights of a convolutional network via a GAN to estimate uncertainty in deep learning predictions, as well as detecting outliers and adversarial examples. Experiment results on semantic segmentation on point clouds, uncertainty estimation and adversarial robustness of CNNs will be shown that significantly improve over prior work. On recurrent networks, 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.


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 has won an NSF CAREER award, (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 video object segmentation, multi-target tracking, point cloud deep networks, uncertainty in deep learning and human understanding of deep learning.


上一篇 没有了