Hi, I'm Zhenghao Xu 徐政豪. Currently, I am a graduate student at Department of Physics, University of California, Santa Barbara. I used to work as a research assistant at Institute for Advanced Study, Tsinghua University. I received my B.S. in Mathematics from School of Mathematical Sciences, Peking University, in July of 2019, major in Applied and Computational Mathematics, as well as minor in Chinese Language & Literature. I worked as game developer in the indie game group MINIBUGS in 2020.
My general research interests cover the interdisciplinary study of Applied Mathematics, Data Science and Astrophysics. Currently, I am working on three projects: (1). High-order Short Characteristic Ray-Tracing Radiation Transfer Simulations. (2) Deep Illustris Inferring Unobservable Structures of Galaxies via Deep Generative Networks (3). Using Unsupervised Deep Learning to disentangle physical properties in Dust-Gas coupled Accretion Diskes.
If you are interested in any problem that I'm working on or related, please contact me! I'm look forward to potential collaborators!
Graduate Student, 2021-Present
Department of Physics, University of California, Santa Barbara
Santa Barbara, California, USA
Researcher Assistant, 2019-2020
Institute for Advanced Study
Tsinghua University, Beijing, China
Advisor: Prof. Xuening Bai
Visiting Student, 2018
Kavli Institute for Astrophysics and Space Research
MIT, Massachusetts, USA
Advisor: Prof. Mark Vogelsberger
Bachelor of Science, 2015-2019
School of Mathematical Sciences
Peking University, Beijing, China
By simply inputting the gas distribution of the galaxy, the GAN(generative adversial network) can unprecedentedly map the dark matter distribution.
MoreLorem ipsum dolor sit amet nisl sed nullam feugiat.
Second-Order Godunov Scheme Finite Volume Method with various approximation Riemann Solvers (Roe, vanLeer, HLLC, etc.), high-accuracy and high-performance.
Multi-Grid Preconditioner Conjugate Gradient Numerical PDE Solver, using both Finite Different Scheme and Finite Element Scheme, in the best case, will reduce the time complexity from O(n^3/2) to O(n).
With Dual Double Deep Q-Network, the agent is able to profit in real stock market without any prior human knowledge.
After self-supervised training with unlabeled font images, the network learns the in-situ distribution of calligraphy and transferring one style to another.
If you are interested in any problem that I'm working on or related, please contact me! I'm look forward to potential collaborators!