Hi, I have interest in how deep learning generalizes in overparameterized regime. My main point of attack is via information geometry, RG flow, and sloppiness.
started my academic life as Undergraduate physicist at Oxford, then Masters at Cambridge. Then I went to work at industrial tech company to live as software engineer and ML researcher in Korea for 5 years. Now I am a DPhil student in Louis group in a search for a theory that explains the peculiar nature of deep learning. You would be surprised how ideas and theories in physics can be used to explain the emergent properties of Deep Learning, and I welcome any discussion on the topic.