Jiyuan Tan bio photo

Jiyuan Tan

I am currently a PhD student in Management Science and Engineering at Stanford University.

  G. Scholar Github e-Mail

My CV

You can also download my full CV.

Research Projects

Neural Causal Partial Identification

joint work with Vasilis Syrgkanis and Jose Blanchet, paper

  • Leveraging the power of generative models, we solve the general partial identification problem with both continuous and discrete variables using neural causal model.
  • We provide detailed construction to the architecture of the neural nets and show the approximation power of Neural Casual Models (NCMs) in theory.
  • We also demonstrate the importance of Lipschitz regularization in the training process and prove that our method is consistency by using Lipschitz regularization.
  • Our method outperforms the previous polynomial programming method.

Zero-Order Optimization

7.2022 -- 7.2023

supervised by Prof. Yinyu Ye

  • Lead a group of three to develop a derivative-free solver for nonlinear constrained optimization in ANSI C. Code available here
  • Use the technique of implicit filtering to increase robustness of algorithm under noises.
  • Combine coordinate search to increase efficiency.

Equilibrium Learning in Offline Markov Game

1.2021 – 6.2021

supervised by Prof. Zhuoran Yang and Prof. Zhaoran Wang

  • Propose the notion of Relative Uncertainty to measure the quality of dataset for offline two-player zero-sum game.
  • Analyze the convergence rate of the proposed algorithm PMVI.
  • Give a lower bound of the convergence rate by constructing a special game.
  • Give a counterexample to illustrate the difference between Markov Decision Process and Markov Game.