My primary research interests are in machine learning, particularly topics at the interface of stochastic dynamical systems and modern neural networks. While there is a focus on using foundational questions to drive advances in the science and engineering of learning systems, the theory is strongly tied to implementational problems and a wide range of very practical applications.
Some central themes of interest:
- Robustness in ML: how to leverage randomness to make ML safer and more reliable
- Sampling & optimization in ML: how to understand and improve sampling and optimization algorithms in ML meaningfully
- ML for sequence modeling: how to use ML to model and learn sequences, particularly those with underlying non-trivial dynamics
- ML + dynamical systems & statistical mechanics: how to use ideas and tools from dynamical systems and statistical mechanics to understand and improve existing ML methods