| layout | page |
|---|---|
| title | About |
I am Johan Sokrates Wind, PhD student at the University of Oslo. My research is mostly concerned with trying to understand the implicit bias in deep learning. Modern neural networks typically have enough capacity to fit the training data many different ways, most of which generalize poorly. I want to understand which configurations of parameters methods like stochastic gradient descent choose, and find what causes these configurations to work so well empirically.