Hi! I'm a first-year CS PhD student studying machine learning at Stanford University, funded by a National Science Foundation Graduate Fellowship. I'm spending my first year rotating with Emma Brunskill, Percy Liang, and Chelsea Finn. I also spend one day a week applying machine learning to systems problems at Google Cloud. Previously, I spent a year as a Google AI resident and before that, completed my master's (CS) and my undergraduate degrees (CS and math) at Stanford, advised by Percy Liang.

I have two main research interests: First, building intelligent agents via reinforcement learning. In particular, I'm interested in building hierarchical agents that can decompose complex tasks into natural and non-degenerate subtasks (either from trial-and-error interaction with the environment or from demonstrations) and can use its learned hierarchies to conduct efficient exploration. In the past, I've explored these ideas on domains such as game-playing, web tasks (e.g., automatically managing your email) and semantic parsing, and I'm excited to extend these ideas to domains such as healthcare and computer security (e.g., automatically detecting vulnerabilities).

Second, applying machine learning to computer systems. Many computer systems (e.g., networking, caching, scheduling) feature increasingly-complex manually-designed and manually-tuned components (e.g., TCP window sizing, cache eviction policies). This paradigm served well in the past, as no viable alternatives existed and manually-engineered components were more easily interpretable and therefore testable. But as these components have grown more complex over time to meet increasing performance requirements, they have become less and less interpretable while performing suboptimally. I am interested in replacing critical previously manually-engineered components of systems with automatically-learned components, which can both yield superior performance and lead to insights into the system. In particular, I've been working on learning cache-eviction policies, which far outperform their manually-engineered counterparts.

Outside of research, you can typically find me climbing or thinking about climbing!



You can reach me at [my first name][my last name]@cs.stanford.edu

Last updated: December 4, 2019