I’m a PhD student at UC Berkeley working with James Demmel, affiliated with the ADEPT, BeBOP, and Theory groups.
I’m broadly interested in making accelerators (both for ML and for other applications) faster, both by mapping computational kernels onto accelerators efficiently and by building tools to help explore the space of accelerator architectures faster. I believe theoretical methods have a lot of promise towards tractably exploring the enormous and often extremely nonconvex search spaces involved, and as a result I’m interested in communication-avoiding algorithms and lower bounds of all types.
If you’d like to collaborate or just say hi, please feel to contact me!