Weijie Su


Prior to joining Penn in Summer 2016, I obtained my Ph.D. from Stanford University in 2016, under the supervision of Emmanuel Cand├Ęs. I received my bachelor's degree in Mathematics from Peking University in 2011. I spent three summers at Microsoft Research (Beijing, 2010; Redmond, 2013; Silicon Valley, 2014).

My Google Scholar profile and CV.


Office: 472 Jon M. Huntsman Hall
3730 Walnut Street
Philadelphia, PA 19104

Email: suw AT wharton DOT upenn DOT edu

Recent news

  • In high-dimensional linear regression, would increasing the signal strengths always improve model selection? Surprisingly, our new paper answers this in the negative in the regime of linear sparsity for the Lasso by introducing a new notion called Effect Size Heterogeneity.

  • Privacy analysis of deep learning using stochastic gradient descent in the GDP framework is now available in TensorFlow. Check out the code and the tutorial for the paper. The maintainer is Zhiqi Bu.

  • I will be offering STAT 991 (Optimization Methods in Machine Learning) this Fall.

  • I received a 2020 Facebook Faculty Research Award.

  • How does the training time of deep learning depend on the learning rate? Check out a new paper, which uncovers a fundamental distinction between nonconvex and convex problems in terms of dependence on the learning rate, showing why learning rate decay is more effective in nonconvex settings. This paper uses a continuous-time surrogate for the analysis of the discrete-time optimization method, just as my earlier Nesterov ODE paper did.

  • I received a 2020 Alfred P. Sloan Research Fellowship.