I work on compilers and systems for machine learning.
I am passionate about designing
easy-to-use compiler systems for compute-efficient deep learning at the edge.
I build compiler and runtime systems for
end-to-end accuracy-aware optimization.
These systems significantly improve the performance and energy usage of deep learning workloads
by systematically trading off small amounts of
computational accuracy, often without impacting the end-to-end task quality.
I also take an interest in designing software debloating frameworks that reduce code size by specializing software programs for user-specified constant configuration settings.