My primary research goal is to make infrastructure software: easy-to-use, fast and reliable. I am generally interested in performance, reliability and programmability in computer systems, esp. ML systems.
Specifically, I am interested in everything related to machine learning compilation such as MLIR and TVM. I am currently building a extensible fuzzer to test any DL framework/compiler with synthesized random yet valid DNNs.
🤗 Feel free to drop me an email if we share common research interest.
ASPLOS’23To AppearFinding Deep-Learning Compilation Bugs with NNSmitharXiv preprint arXiv:2207.13066 2022
OOPSLA’22Coverage-guided tensor compiler fuzzing with joint IR-pass mutationProceedings of the ACM on Programming Languages 6 (OOPSLA1), Apr 2022
ACMMM’21 (OSC)Fast and Flexible Human Pose Estimation with HyperPoseProceedings of the 29th ACM International Conference on Multimedia, Apr 2021
OctoML, Smr. 2022 Pattern Language
TalksFinding Deep-Learning Compilation Bugs with NNSmith
- Software Engineering Retreat, University of Illinois at Urbana-Champaign Sept. 2022