Yijun Wan

I am currently a Postdoctoral Researcher in game theory at CEREMADE, Université Paris Dauphine - PSL, under the supervision of Bruno Ziliotto and Joon Kwon.
Previously, I completed my PhD thesis under the supervision of Dmitry Chelkak, focusing on the intersection of probability, statistical mechanics, and complex analysis.

You can download my Curriculum Vitae here.

Email: yijun.wan at normalesup dot org


Papers

In Probability:

[1] On the convergence of massive loop-erased random walks to massive SLE(2) curves, with Dmitry Chelkak, Electron. J. Probab., 2021
[2] Capacity of the range of tree-indexed random walk, with Tianyi Bai, Ann. Appl. Probab., 2022
[3] On the crossing estimates of simple conformal loops ensembles, with Tianyi Bai, Int. Math. Res. Not., 2022.

In Machine Learning:

[1] Chaotic Regularization and Heavy-Tailed Limits for Deterministic Gradient Descent, S.H. Lim, Y. Wan, U. Simsekli, NeurIPS, 2022.
[2] Implicit Compressibility of Overparametrized Neural Networks Trained with Heavy-Tailed SGD, Y. Wan, M. Barsbey, A. Zaidi and U. Simsekli, ICML, 2024.
[3] Lessons from Generalization Error Analysis of Federated Learning: You May Communicate Less Often, M. Sefidgaran, R. Chor, A. Zaidi, Y. Wan, ICML, 2024.

In Game Theory:

[1] Time-Dependent Blackwell Approachability and Application to Absorbing Games, with Joon Kwon, Bruno Ziliotto. arXiv:2303.04956, 2024.

Other Scientific Writings:

PhD thesis: Statistical Mechanics Models via the Lens of Potential Theory, defended on 26 September 2022, with 3 published Chapters (3, 4 and 5) and unpublished Chapter 6 on near-critical Ising and dimer models. You can find the slides here.
Mémoire de M2: Schramm-Loewner Evolutions and Conformal Loop Ensembles. You can find the report here.
Mémoire de la première année: Domono tilings and the Gaussian free field, with Valentin Schmutz. You can find the report here.
MVA RL Project: UCRL with post-decision state: regret analysis. You can find the report and the slides here.