Xingyou Song
Research Scientist at Google DeepMind
United States
Summary
🤩
Rockstar🎓
Top SchoolXingyou Song is a Senior Research Scientist at Google DeepMind with 11 years of experience applying ML research to large-scale systems, specializing in LLMs for optimization, data science, and reasoning. Previously on Google Brain and an OpenAI research resident, he has deep expertise in automated machine learning, reinforcement learning, meta-learning, and efficient attention mechanisms. His open-source contributions include core work on Google Vizier—implementing proto converters and parameter configuration for black-box/hyperparameter optimization—and implementations of ES-MAML and fast self-attention variants in the Google Research repo. Comfortable moving between research and production engineering, he blends algorithmic rigor with practical system design to ship reproducible ML tooling. Trained at UC Berkeley, he brings experience from top research labs and quant internships, giving him a rare mix of theoretical depth and market-facing quantitative engineering.
11 years of coding experience
4 years of employment as a software developer
Bachelor's degree, Computer Science, Bachelor's degree, Computer Science at University of California, Berkeley
Github Skills (21)
algorithm10
pytorch10
python10
protocol-buffers10
attention-mechanism10
machine-learning10
parameterized10
ai10
params10
para10
jax10
grpc10
numpy9
hyperparameter-optimization9
data-structures8
Programming languages (5)
Github contributions (5)
google/vizier
Mar 2022 - Jan 2023
Python-based research interface for blackbox and hyperparameter optimization, based on the internal Google Vizier Service.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:52 releases, 264 commits, 30 PRs in 11 months
Contributions summary:Xingyou's contributions primarily involve developing the initial version of the Vizier code, focusing on proto conversion and parameter configuration management. The user created converters for OSS Vizier's protos and implemented parameter configurations, including bounds, feasible points, and child parameter configurations. Moreover, the user updated existing comments and README files, refactoring the GRPC server, and adding new functionalities for suggesting trials.
hyperparameter-optimizationpythonviziertuningtuning-parameters
google-research/google-research
Jul 2020 - Nov 2021
Google Research
Role in this project:ML Engineer
ML EngineerContributions:1 review, 43 commits, 6 comments in 1 year 4 months
Contributions summary:Xingyou primarily contributed to the `es_maml` and `performer` directories, indicating involvement in meta-learning and attention mechanism research. They implemented the ES-MAML algorithm, fixed bugs, and updated the README. Furthermore, they contributed to the development of "Fast Self Attention" and "SLiMPerformer" models, including unidirectional variants and code optimizations, with a focus on fast softmax and generalized attention mechanisms. These contributions span multiple papers and implementations, focusing on core AI/ML tasks.
googlemachine-learningai
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.