Welcome to the lab of Machine Learning Stochastic Physics! Our research topics include stochastic process, machine learning, nonequilibrium statistical physics. Combining analytical and numerical approaches, the ultimate goal is to uncover simple theoretical principles that could help understand complex dynamical processes and design efficient machine learning algorithms.
To discuss any interesting science, feel free to contact us at jamestang23@gmail.com
About PI: Since 2024, I am a Professor at Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu, my hometown. From 2021 to 2024, I was an Associate Researcher at International Academic Center of Complex Systems, Beijing Normal University, Zhuhai. From 2018 to 2021, I was a Postdoc fellow at Signaling Systems Laboratory in University of California, Los Angeles, supported by Prof. Alexander Hoffmann. I obtained my Ph.D. from Department of Physics, Shanghai Jiao Tong University in 2018, mentored by Prof. Ping Ao. During 2016 to 2018, I was an exchange graduate student in Department of Physics, University of California, San Diego, where I got quantitative training from Prof. Terence Hwa. I completed my B.S. in honored class of Zhiyuan College, Shanghai Jiao Tong University in 2013.
Research interests
Machine learning stochastic dynamics:
Machine learning and stochastic dynamics have deep connections and cross-feed each other. We recenetly have developed machine-learning approaches to investigate the time evolution of stochastic dynamics: (1) propose the first approach of using the neural network alone to solve the chemical master equation; (2) characterize a type of dynamical phase transition in nonequilibrium statistical mechanics; (3) learning noise-induced transitions by multi-scaling reservoir computing.
Dynamical information theory:
Inferring mutual information from time series data remained challenging as the possible trajectory configurations increases exponentially with the number of time points. We develop a computational framework to quantify the dynamical mutual information of intracellular signaling process, and summarize the recent progresses in this review on quantifying information by machine learning.
Nonequilibrium and quantum thermodynamics:
As a remarkable advance in nonequilibrium thermodynamics during the last 20 years, Jarzynski equality connects free energy changes to nonequilibrium work fluctuations. We found that the free energy change through the Jarzynski equality is independent of magnetic field in the classical regime, but can be amplified by magnetic field in driven quantum system. The magnetic field can also be generated by the coupling to the heat bath.
Stochastic process without detailed balance:
Stochastic transitions are ubiquitous in nature. Based on path integral approach, we develop a scalable numerical approach to calculate transition rates for a class of Langevin dynamics. The computational cost is robust to varying noise intensity, beyond small noise limit. The efficient computations on transition rates enable a broader use of stochastic modeling in complex dynamics, such as cell state transitions.
Quantitative biology:
To understand the complex behaviors in biological systems across scales, we have attempted to construct minimum models to describe the data and make predictions: (a) we identified computationally a molecular circuit that control necroptosis decisions such that a bimodal death-time distribution can be produced; (b) we demonstrated that chemotaxis in nutrient-replete conditions promotes the expansion of bacterial populations by modeling the physiological effect on bacterial chemotaxis.
Join us
Postdocs
We are looking for motivated postdocs in areas of machine learning, stochastic dynamics, and statistical physics. Research topics include but are not limited to tracking time evolution and phase transitions in stochastic and open quantum systems by neural networks. The Institute of Fundamental and Frontier Sciences focuses on high-level interdisciplinary science and has an international environment with a dozen foreign faculties. Applicants should have a Ph.D. in physics, math or related subjects. Please send me a CV and a research statement. For more information.
Students
We are always looking for self-driven Ph.D. students, master, and undergraduates. If you are interested in, please get in touch. For more information.