Zheng XU 许正

xuzhustc AT gmail  [Curriculum Vitae][Google Scholar]

 Zheng's portrait

Welcome to Zheng's site! I am a graduate student in the department of Computer Science at the University of Maryland, College Park. I am currently working on optimization, machine learning and their applications, under the supervision of Dr.
 Tom Goldstein.

My recent interests are: adaptive stepsize in optimization, primal-dual method in optimization, neural network acceleration, generative adversarial networks for vision and language, and distributed optimization.

I am also interested in some topics I have previously worked with: low rank and sparse models, domain adaptation in computer vision, topic models and tensor methods, 
NER and information extraction, image retrieval, and mining information from large-scale multimedia data.  

Before joining Maryland, I was a Project Officer at Visual Computing Research Group, School of Computer EngineeringNanyang Technological University, Singapore.
 I received my M.Eng. and B.Eng. in the Department of Electronic Engineering and Information Science, the University of Science and Technology of China (USTC), advised by Dr. Chang Wen Chen, Dr. Bin Liu and Dr. Houqiang Li. During my study as a master student, I was fortunate to take a long-term internship  at  Microsoft Research Asia (MSRA) with Dr. Xin-Jing Wang.

I have interned/collaborated with companies: Apple, Honda, Amazon, IBM, Rolls-Royce and Microsoft.


  • Sep 7 release preliminary version of PyTorch code for the image dehazing work at BMVC 2018. Stay tuned.
  • Sep 7, "Visualizing the Loss Landscape of Neural Nets "  has been accepted to NIPS 2018  (acceptance rate 20.8%), congrats Hao and all coauthors. 
  • Sep 3 - 6, travel to BMVC, present GAN for accelerating NN and dehazing
  • May 21 - Aug 24, intern with Apple. We anonymously submitted a draft to a workshop.
  • Jul 7, two papers are accepted to BMVC 2018 (acceptance rate 29.9%), thanks all coauthors. GAN for knowledge distillation was my intern project at Honda last summer. Dehazing baseline achieves impressive performance with a simple yet effective method.
  • Jan 8 - Mar 30, intern with Adobe at San Jose. The project is about image style transfer. We have anonymously submitted our draft to a conference . 
  • Mar 17, released part of the code for our ACADMM paper (ICML'17). Have fun!
  • Mar 15, " Learning to Cluster for Proposal-Free Instance Segmentation" has been accepted to IJCNN 2018, congrats to Yen-Chang, and all coauthors.
  • Jan 29, "Stabilizing Adversarial Nets with Predictive Methods" has been accepted to ICLR 2018 (acceptance rate 34%), congrats to Sohil and Abhay, and all coauthors.
  • Dec 4-9, travel to NIPS@Long Beach; poster on "Training Quantized Nets: A Deeper Understanding" at main conference; poster on "Visualizing the Loss Landscape of Neural Nets " at workshop Deep Learning: Bridging Theory and Practice.
  • Nov 29, passed proposal exam, and advanced to candidacy.
  • Nov 7, "Towards Perceptual Image Dehazing by Physics-based Disentanglement and Adversarial Training" has been accepted to AAAI 2018 (acceptance rate 25%), congrats Xitong and all coauthors.
  • Sep 4, "Training Quantized Nets: A Deeper Understanding" has been accepted to NIPS 2017 (acceptance rate 21%), congrats Hao and Soham, and all coauthors.
  • Older news