Altan Haan
Senior AI Compiler Engineer, MLIR at NVIDIA
Berkeley, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAltan Haan is a Senior AI Compiler Engineer specializing in MLIR with a decade of experience spanning compilers, program analysis, and deep learning frameworks. He combines industry impact—working on deep learning compiler optimizations at OctoML (acquired by NVIDIA) and now at NVIDIA—with three years of Berkeley research in compilers, HPC, and query optimization during a paused PhD. Practically minded, he contributed to TVM’s Relay backend, improving automatic differentiation and training primitives (including gradients for pooling and concatenate operators) and adding memory-optimization checkpointing and tests. Based in Berkeley, he bridges PL and ML systems, focusing on making training workloads faster and more robust across CPUs, GPUs, and accelerators. A detail-oriented engineer, he often surfaces subtle correctness issues in gradients and reduces memory footprints while keeping performance scalable.
10 years of coding experience
3 years of employment as a software developer
Doctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at UC Berkeley Electrical Engineering & Computer Sciences (EECS)Bachelor of Science - BS, Computer Science, Bachelor of Science - BS, Computer Science at University of WashingtonEarly Entrance Program, Early Entrance Program at UW Robinson Center
Chinese
Github Skills (12)
tensorrt10
tvm10
automatic-differentiation10
machine-learning10
tensor10
deep-learning10
tensorflow10
operation10
python10
gradient10
compiler-compiler9
compiler9
Programming languages (4)
Github contributions (5)
apache/tvm
Jun 2019 - Jun 2022
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Role in this project:
Backend Developer & ML Engineer
Backend Developer & ML EngineerContributions:118 reviews, 31 commits, 32 PRs in 3 years
Contributions summary:Altan contributed to the development of the Relay compiler within the TVM project, focusing on enhancements related to automatic differentiation (AD) and training features. Their contributions included bug fixes for reduce axis functionality, adding and improving gradients for various operators such as `max_pool2d`, `global_avg_pool2d`, `concatenate`, and others. They also implemented checkpoint annotation for memory optimization and added tests to ensure the correctness of these gradient calculations.
metalvulkancompilertensoropencl
altanh/colocolo
Feb 2019 - Mar 2019
A synthesis-enabled solver for relational logic
Contributions:40 commits, 31 pushes in 23 days
relationalsynthesissolverrelational-algebragrammars
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.