Dhruva Karkada

Personal Website

Reveal Contact Info

PhD Student

Physics

DeWeese Lab

Current Research

I’m interested in the learning mechanisms of deep neural networks. Their learning dynamics offer a scientifically-accessible playground for understanding feature learning and reasoning. I use tools from applied math and statistical physics to develop quantitative understanding of these systems, and I’m especially interested knowing why deep learning is often much more sample-efficient than classical machine learning techniques.

Background

Before this, I studied physics, astronomy, and computer science at UT Austin. During the pandemic I went down a Wikipedia rabbithole trying to learn “what’s going on in my brain” and was immediately confronted by the sheer complexity of neural systems. I wondered “is there any hope of ever understanding (in a physics sense) how learning happens?” I’ve been working on this problem ever since. In my free time, I enjoy cooking for friends, playing chess, messing around with synthesizers, watercoloring, and going on long walks.