About Me

I am currently a principal scientist in TuSimple. I got my PhD degree from CSE department, HongKong University of Science and Technology in 2015. My supervisor is Prof. Dit-Yan Yeung. Before that, I got my BS degree from Zhejiang University, 2011 under the supervision of Prof. Zhihua Zhang.

I am recruiting self-motivated interns in computer vision and data mining. If you are interested in, please send your CV to my email.

Research Interest

My research interest focuses on applying statistical computational model to real problems in computer vision and data mining. Currently, I mainly work on matrix factorization and deep learning. Especially I am interested in the area of visual tracking, object detection, image classification and recommender system.

Experience

  • Nov. 2010 - Apr. 2011, Intern at Microsoft Research Asia, Media Computing Group. Work with Dr. Jingdong Wang.
  • May. 2014 - Aug. 2014, Intern at Baidu, Institute of Deep Learning. Work with Dr. Yinan Yu.
  • Aug. 2014 - Feb. 2015, Visiting scholar at Carnegie Mellon University, Robotics Institute. Work with Dr. Abhinav Gupta.

Updates

  • The project page for " Understanding and Diagnosing Visual Tracking Systems" (ICCV15') is set up. The codes are available as well. We analyzed the important factors that contributes to a good performance in a complicated visual tracking system. Then we proposed a simple baseline that only uses HOG + logistic regression, yet its performance can beat most of the papers. (10/17/2015)
  • A paper on analysis of different rectified activation functions has been accepted by ICML DL Workshop 15'. We show that leaky ReLU and its variants could consistently outperform ReLU. Simple but effective modification to improve the performance!
  • A paper on applying deep learning to content based recommender system has been accepted by SIGKDD15'. Dramastic improvement over LDA based system (KDD 12' best student paper)!
  • Two papers on robust matrix factorization and multimedia event detection/recounting are accepted at CVPR15'. Final versions are available.