Geoff Pleiss
Research Scientist at University of British Columbia
Vancouver, British Columbia, Canada
Summary
🤩
RockstarGeoff Pleiss is a research scientist and assistant professor-level statistician based in Vancouver with 14 years of software and ML engineering experience. He bridges rigorous probabilistic modeling and practical engineering, contributing notable open-source work such as batch-mode multitask extensions to the widely used gpytorch Gaussian process library and memory-efficient DenseNet implementations optimized for CUDA. Geoff is equally comfortable improving developer-facing UI components—adding quality and ergonomics to a major design system—as he is optimizing backend ML pipelines, reflecting a rare full-stack fluency in research software. His background suggests a focus on scalable, memory- and compute-efficient solutions for real-world ML problems, with an emphasis on reproducible, open-source tooling.
14 years of coding experience
Github Skills (22)
pytorch10
javascript10
python10
machine-learning10
user-interface10
system-design10
carbon-design-system10
gaussian-processes10
system10
densenet10
component-library10
deep-learning10
sys10
cuda10
react10
Programming languages (11)
Github contributions (5)
gpleiss/efficient_densenet_pytorch
May 2017 - Feb 2021
A memory-efficient implementation of DenseNets
Role in this project:
ML Engineer
ML EngineerContributions:1 review, 74 commits, 16 PRs in 3 years 9 months
Contributions summary:Geoff contributed to the implementation of memory-efficient techniques within a DenseNet architecture. Their work involved optimizing the model's performance through the use of shared memory allocations, efficient concatenation, and CUDA-optimized batch normalization. They introduced several modifications to the existing DenseNet implementation, including the creation of an efficient bottleneck layer and the use of checkpointing to enhance memory usage. The user's changes focused on improving the efficiency and memory footprint of the deep learning model.
efficientnetmemoryvisual-recognitionmemory-efficientdeep-learning
cornellius-gp/gpytorch
Jun 2017 - Jan 2023
A highly efficient implementation of Gaussian Processes in PyTorch
Role in this project:Back-end & ML Engineer
Back-end & ML EngineerContributions:32 releases, 208 reviews, 1442 commits in 5 years 8 months
Contributions summary:Geoff implemented batch mode functionality for the multitask Gaussian Processes by modifying the existing code for a given test example. Their modifications include changes to the code across multiple files to enable the use of a batch of datasets. The code changes suggest the addition of new functionalities for dealing with multiple, possibly heteroskedastic, data sets.
pytorchgpu-accelerationgaussianstochastic-processesmodular
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.