Marc Brockschmidt
Senior Staff Research Scientist at Google DeepMind
Cambridge, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolMarc Brockschmidt is a Senior Staff Research Scientist at Google DeepMind in Cambridge with 12 years’ experience applying deep learning to highly structured data such as code and molecules. He holds a PhD from RWTH Aachen and marries deep expertise in program analysis and verification with practical ML engineering—turning formal techniques into developer-facing models. At Microsoft he led research on program synthesis, termination and complexity analysis while also optimizing graph neural networks, contributing performance and feature improvements (residual connections, edge-wise attention, sparse-GNN optimizations) to Microsoft’s GNN sample repos. Today he focuses on teaching machines to assist software developers, known for pragmatic optimizations that make research models faster, more flexible and easier to reproduce.
12 years of coding experience
11 years of employment as a software developer
Doctor of Philosophy (PhD) Computer Science, Doctor of Philosophy (PhD) Computer Science at RWTH Aachen University
German, English
Github Skills (15)
model-building10
machine-learning10
tensorflow10
graph-neural-network10
performance-optimization10
python10
modeling10
model-driven10
model-driven-development10
trainings9
subsampling8
benchmark8
benchmarking8
resampling8
sampling8
Programming languages (10)
Github contributions (5)
microsoft/gated-graph-neural-network-samples
Nov 2017 - Sep 2019
Sample Code for Gated Graph Neural Networks
Role in this project:
ML Engineer
ML EngineerContributions:34 commits, 7 PRs, 18 pushes in 1 year 10 months
Contributions summary:Marc primarily focused on optimizing and enhancing the performance of Gated Graph Neural Network models within the repository. Their contributions include refactoring code, adding new functionalities like residual connections and edge-wise attention, and implementing various optimizations to improve the speed and efficiency of the models, especially related to sparse GNNs. The user also introduced several configurable parameters to the model, allowing for greater flexibility, including allowing the user to simulate graph convolutional networks. Furthermore, the user integrated features for saving, restoring, and sampling examples for several tasks.
dataminingneural-graphneural-networksgraph-neural-networkgraph-neural-networks
microsoft/tf-gnn-samples
Jun 2019 - Apr 2021
TensorFlow implementations of Graph Neural Networks
Role in this project:ML Engineer
ML EngineerContributions:34 commits, 5 PRs, 16 pushes in 1 year 9 months
Contributions summary:Marc primarily contributed to implementing and improving Graph Neural Network (GNN) models within the TensorFlow framework. They added new GNN variants, specifically an edge-MLP based R-GCN model and made improvements to existing models like the GIN. Furthermore, the user worked on optimizing the training process by adding features such as normalized learning rates and better logging for multiple runs. They also contributed to the benchmark scripts and documented the results.
implementationsdeep-learningneural-graphneural-networksgraph-neural-networks
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.