Project Overview#

Optimizers are essential for model training in machine learning. However, there lacks a way to estimate their performance on out-of-distribution data. For example, would a strong in-distribution optimizer work well on out-of-distribution datasets?

This project will investigate possible ways to resolve this problem and perhaps improves model robustness in the long run.

Requirements#

  • Knowledge of using Pytorch for deep learning studies.
  • Strong math abilities.

Info#

bars search times arrow-up