Michael Anderson
Senior Deep Learning Architect at NVIDIA
Madison, Wisconsin, United States
Summary
🤩
Rockstar🎓
Top SchoolMichael Anderson is a Senior Deep Learning Architect with a decade of experience bridging research and production at top labs including NVIDIA, Meta/Facebook, and Intel Labs. He holds a PhD from UC Berkeley and builds high-performance ML infrastructure—his open-source work on PyText, FBGEMM, and Glow focuses on compiler integrations, jagged-tensor operators, and quantization tooling that accelerate real-world model deployment on specialized hardware. Known for deep performance profiling and microbenchmarking, he has delivered kernel-level optimizations and prototype integrations (e.g., NNPI, NVIDIA jagged operators) that measurably improve GFLOPS and throughput. Based in Madison, WI, he blends academic rigor with systems engineering to turn advanced research into production-grade inference stacks.
10 years of coding experience
9 years of employment as a software developer
PhD Electrical and Computer Engineering, PhD Electrical and Computer Engineering at University of California, BerkeleyBS Computer Engineering, BS Computer Engineering at University of Wisconsin-Madison
Github Skills (28)
code-optimization10
pytorch10
quants10
benchmark10
c-language10
fba10
emm10
operation10
python10
tensorrt10
benchmarking10
hardware-acceleration10
f10
tensorflow10
performance-optimization10
Programming languages (3)
Github contributions (5)
pytorch/glow
Oct 2019 - Sep 2021
Compiler for Neural Network hardware accelerators
Role in this project:
Backend Developer & Performance Engineer
Backend Developer & Performance EngineerContributions:81 commits, 80 PRs, 3 comments in 1 year 11 months
Contributions summary:Michael focused on performance microbenchmarks within the Glow compiler for neural network hardware accelerators. Their contributions included creating and modifying microbenchmarks for GEMM, addition, SLS (Sparse Lengths Weighted Sum), batch GEMM, and transpose operations. They implemented and optimized these benchmarks, with a focus on measuring GFLOPS and GBytes/sec, indicating a strong emphasis on performance profiling and code optimization within the Glow framework. They also made contributions related to adding new kernels and features to the project.
hardware-acceleratorscompilerneural-networkacceleratorshardware
pytorch/FBGEMM
Mar 2022 - Dec 2022
FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/
Role in this project:ML Engineer
ML EngineerContributions:3 releases, 2 reviews, 32 commits in 8 months
Contributions summary:Michael's contributions primarily involve implementing and optimizing operations within the fbgemm library, particularly focusing on jagged tensor manipulations. The user added a new operator, `dense_to_jagged`, which converts a dense tensor into a jagged tensor, and also refactored existing jagged tensor operations, improving performance by converting to dense tensors for elementwise operations. Further work included debugging and benchmarking the new jagged tensor operations. They also integrated a prototype jagged operator from NVIDIA and added a benchmark.
matrix-multiplicationfacebookmultiplicationmatrixml-applications
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.