Research & Publications

Research Overview

I am interested in developing provable algorithms and lower bounds for simple structures or phenomena that explain learning in interactive, adaptive, and modern machine learning systems. To be more precise:

• Online learning & learning in adaptive environments. I am interested in online learning, learning dynamics in games, reinforcement learning.
• Deep learning theory. I am interested in optimization and learning dynamics, model expressivity, and generative models such as diffusion models.
• Sampling & statistical estimation. I am interested in a lot of algorithmic statistics related question, especially sampling and robustness.
• Classical learning theory. I am interested in classical PAC learning theory or related style frameworks about sample complexity of learning, active learning, etc.

Across these areas, I tend to gravitate toward clean theoretical questions with an extremal or optimization flavor: sharp thresholds, minimax rates, optimality guarantees and lower bounds.

You can also find my publications on my Google Scholar profile.

If you are interested in collaborating, feel free to reach out at viverson@mit.edu.

Publications & Preprints

★ denotes representative work.

Talks