About

I am an assistant professor of Institute for Theoretical Computer Science at SUFE. Before joining SUFE, I did my postdoc at SUTD under supervision of Georgios Piliouras and Ioannis Panageas. I got my B.S. from China University of Geosciences (Beijing) and Ph.D in Mathematics from University at Buffalo, advised by Mohan Ramachandran.

Publication

• On the Last-iterate Convergence in Time-varying Zero-sum Games: Extra Gradient Succeeds where Optimism Fails
  with Yi Feng, Hu Fu, Qun Hu, Ping Li, Ioannis Panageas and Bo Peng,
  NeurIPS 2023.


• Embedding Surfaces by Optimizing Neural Networks with Prescribed Riemannian Metric and Beyond,
  with Yi Feng, Sizhe Li, and Ioannis Panageas,
  ICML 2023 Workshop Frontiers4LCD.


• Semi Bandit dynamics in Congestion Games: Convergence to Nash Equilibrium and No-Regret Guarantees,
  with Ioannis Panageas, Efstratios Panteleimon Skoulakis, Luca Viano and Volkan Cevher,
  ICML 2023 (Oral).


• AdaGrad Avoids Saddle Points,
  with Kimon Antonakopoulos, Panayotis Mertikopoulos and Georgios Piliouras,
  ICML 2022.


• Accelerated Multiplicative Weights Update Avoids Saddle Points almost always,
  with Yi Feng and Ioannis Panageas,
  IJCAI 2022.


• Constants of Motion: The Antidote to Chaos in Optimization and Game Dynamics,
  with Georgios Piliouras,
  arXiv.


• Last iterate convergence in no-regret learning: constrained min-max optimization for convex-concave landscapes,
  with Qi Lei, Sai Ganesh Nagarajan and Ioannis Panageas,
  AISTATS 2021.


• Fast convergence of Langevin dynamics on manifold: Geodesics meet log-Sobolev,
  with Qi Lei and Ioannis Panageas,
  NeurIPS 2020.


• Convergence to second-order stationarity for non-negative matrix factorization: Provably and concurrently,
  with Ioannis Panageas, Stratis Skoulakis and Antonis Varvitsiotis,
  arXiv.


• Depth-width trade-offs for ReLU networks via Sharkovsky's theorem,
  with Vaggos Chatziafratis, Sai Ganesh Nagarajan and Ioannis Panageas,
  ICLR 2020 (spotlight), Video (Speaker: Ioannis Panageas, MiFODS Workshop, MIT, Jan 27-29, 2020).


• First-order methods almost always avoid saddle points: the case of vanishing stepsize,
  with Ioannis Panageas and Georgios Piliouras,
  NeurIPS 2019, Poster.


• Multiplicative weights update as a distributed constrained optimization algorithm:
  converges to second-order stationary points almost always,
  with Ioannis Panageas and Georgios Piliouras,
  ICML 2019, Poster, Slides (Speaker: Xiao Wang).


• Kazhdan's property (T) and structure for locally symmetric spaces,
  Submitted