About me

Hi there! This is Qingyao Sun, a MS student in Statistics at the University of Chicago. My research interest lies in finding classical optimization solutions to modern ML problems, and vice versa. For example, some problems I want to explore include:

• Optimization → Machine Learning
  • Can we practically achieve superlinear convergence in neural network training, possibly by following OASIS or SuperPolyak?
  • How do these optimizers affect the generalization and robustness of the model?
• Optimization ← Machine Learning
  • What is a good algorithm for large-scale nonlinear programming on AI accelerators like GPUs and TPUs?
  • How do momentum-based methods like Nesterov Accelerated Gradient work for non-convex optimization?

Ultimately, I aim towards bridging different “tribes” in academia so that everyone gets to play in everyone else’s backyard.