Ian Osband
Research Scientist at Google
London, England, United Kingdom
Summary
👤
Senior🎓
Top SchoolIan Osband is a research scientist with 11 years of experience building and operationalizing reinforcement learning and decision-making systems at DeepMind, OpenAI and now Google. He combines rigorous academic training (PhD Stanford, MMath Oxford) with hands-on engineering—contributions to influential open-source RL projects like DeepMind’s bsuite and Acme show deep expertise in exploration, replay mechanisms, and scalable agent design. His work spans foundational research (epistemic neural networks, efficient agents) and product-focused research powering systems like ChatGPT and O-series. Ian’s background in quantitative finance and large-scale metrics gives him a practical bent for pressure-tested solutions and real-world impact. Colleagues describe him as someone who rethinks uncertainty in deep learning, turning theoretical insight into robust, deployable algorithms. Based in London, he blends curiosity-driven research with a track record of shipping high-impact ML systems.
11 years of coding experience
10 years of employment as a software developer
MMath Mathematics, MMath Mathematics at University of OxfordKing's Scholar, King's Scholar at Eton CollegeDoctor of Philosophy (PhD) Management Science and Engineering, Doctor of Philosophy (PhD) Management Science and Engineering at Stanford University
Github Skills (13)
dqn10
machine-learning10
agent10
jax10
python10
reinforcement-learning10
optim9
reverse9
data-analysis9
opt9
pandas9
haiku8
numpy8
Programming languages (2)
Github contributions (5)
google-deepmind/bsuite
Aug 2019 - Jun 2022
bsuite is a collection of carefully-designed experiments that investigate core capabilities of a reinforcement learning (RL) agent
Role in this project:
ML Engineer
ML EngineerContributions:25 commits, 33 comments, 12 issues in 2 years 10 months
Contributions summary:Ian primarily contributed to the analysis and experimentation aspects of the `bsuite` repository, which focuses on reinforcement learning (RL) agent capabilities. Their work involved modifying and extending existing analysis code, including making the number of episodes configurable, and adjusting scoring methods. They also made changes to the core RL environment code, making the noise come from the seed RNG, and fixed bugs.
stable-baselinesagentreinforcement-learningdeep-reinforcement-learningcapabilities
google-deepmind/acme
Nov 2020 - Dec 2020
A library of reinforcement learning components and agents
Role in this project:ML Engineer
ML EngineerContributions:6 commits in 13 days
Contributions summary:Ian primarily contributes to the development and refinement of reinforcement learning agents within the Acme library. Their work involves refactoring existing DQN learner components, separating SGD logic for easier configuration, and integrating prioritized replay mechanisms. They also focus on configuring and optimizing replay buffers, and defining loss functions, demonstrating a deep understanding of reinforcement learning algorithms. Further, the user introduces replay logic abstractions, indicating efforts to improve modularity and shareability across various agent implementations.
reinforcement-learningreinforcementagentsdeep-reinforcement-learning
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.