Hi, I’m Victor (Xiao-Jie) Zhang, currently an employee @NVIDIA. I’m currently working as a Deep Learning Architect, who studies possible methods to speedup training/inference for DL models on GPU platforms. My interests are deep learning models and their various applications in many fields. I’m familiar with deep learning frameworks such as TensorFlow and PyTorch, as well as their underlying deep learning kernels written in CUDA/C++.

My past project experience includes the application of ExMy floating point representation (i.e. x exponent digits and y mantissa digits) in deep learning training/inference; Int8 quantization inference for deep learning models etc. And I’ve also had the experience of writing kernels for deep learning operators.

I’ve acquired my doctorate degree in Fudan University, in the field of theoretical chemistry, so you can expect many of my github repos are related to this field.

Following is my CV and publication list.

  • Fudan University 2012.09—2017.06

    Degree: Ph.D. of Theoretical Chemistry

    (DFT simulation of chemical reactions, High performance computing using MPI and CUDA)

  • Fudan University 2012.09—2017.06

    Degree: B.S. of Chemistry

  • NVIDIA Semiconductor Technology (Shanghai) Co., Ltd. 2018.04—Now

    Senior Deep Learning Architect

    Working on deep learning models with reduced precision and sparsity as well as their CUDA optimization.

  • Hujiang Education & Technology (Shanghai) Co. Ltd. 2017.05—2018.04

    AI Engineer

    Developing and deploying deep learning natural language processing (NLP) and image processing models.

(As Ph.D. student)

  1. From Atoms to Fullerene: Stochastic Surface Walking Solution for Automated Structure Prediction of Complex Material, J. Chem. Theory Comput., 2013, 9, 3252
  2. Double-Ended Surface Walking Method for Pathway Building and Transition State Location of Complex Reactions, J. Chem. Theory Comput, 2013, 9, 5745
  3. Reaction Sampling and Reactivity Prediction Using Stochastic Surface Walking Method，Phys. Chem. Chem. Phys., 2015, 17, 2757
  4. Variable-Cell Double-Ended Surface Walking Method for Fast Transition State Location of Solid Phase Transition, J. Chem. Theory Comput, 2015, 11, 4885
  5. Pressure-induced Silica Quartz Amorphization Studied by Iterative Stochastic Surface Walking Reaction Sampling, Phys. Chem. Chem. Phys., 2017, 19, 4725
  6. Stochastic surface walking reaction sampling for resolving heterogeneous catalytic reaction network: A revisit to the mechanism of water-gas shift reaction on Cu, J. Chem. Phys., 2017, 147, 152706
  7. Stochastic Surface Walking Method for Crystal Structure and Phase Transition Pathway Prediction, Phys. Chem. Chem. Phys., 2014, 16, 17845
  8. Energy Landscape of Zirconia Phase Transitions, J. Am. Chem. Soc., 2015, 137, 8010
  9. Graphite to Diamond: Origin for Kinetics Selectivity, J. Am. Chem. Soc., 2017, 139, 2545

(As NVIDIA’s employee)

  1. Optimizing Multi-GPU Parallelization Strategies for Deep Learning Training, IEEE Micro, 2019, 39, 5 (by name Victor Zhang)
  2. Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation, arXiv: 2004.09602v1 (by name Xiaojie Zhang)