I am an AI scientist at Prior Labs in Berlin, where I build tabular foundation models: tools that understand spreadsheet-structured data, and combine it with common sense, domain knowledge, or other data, to make high-quality predictions. I am also finishing my PhD at UCL in the Centre for Artificial Intelligence, supervised by François-Xavier Briol and Marc Deisenroth.

My main interests are making large machine learning models more efficient, via both modelling and systems approaches, and writing nice software along the way. I'm primarily excited about applications of AI to science.

Bio

2025 -

AI Scientist Prior Labs

2020 -

PhD candidate in Computer Science UCL Supervised by François-Xavier Briol and Marc Deisenroth Internships: Graphcore (implementing efficient sparse attention in Transformers), Seqana (measuring soil carbon content using satellite imagery)

2019 - 2020

Research Assistant in Bayesian Deep Learning University of Oxford OATML group, supervised by Yarin Gal

2018 - 2019

MSc in Computer Science University of Oxford

2016 - 2018

Software Engineer Google Encryption of Android backup data using a client-side key, released in Android 9

2013 - 2016

BA in Computer Science University of Cambridge

Research and code

• Approximate Top-k for Increased Parallelism Oscar Key, Luka Ribar, Alberto Cattaneo, Luke Hudlass-Galley, Douglas Orr arXiv PyTorch/CUDA impl. ENLSP workshop NeurIPS 2024
• Scalable Data Assimilation with Message Passing Oscar Key*, So Takao*, Daniel Giles*, Marc Peter Deisenroth arXiv code Climate Informatics 2024 oral
• No Train No Gain: Revisiting Efficient Training Algorithms For Transformer-based Language Models Jean Kaddour*, Oscar Key*, Piotr Nawrot, Pasquale Minervini, Matt J. Kusner arXiv code NeurIPS 2023 poster
• Interlocking Backpropagation: Improving depthwise model-parallelism Aidan N. Gomez*, Oscar Key*, Kuba Perlin, Stephen Gou, Nick Frosst, Jeff Dean, Yarin Gal arXiv code JMLR
• Optimally-Weighted Estimators of the Maximum Mean Discrepancy for Likelihood-Free Inference Ayush Bharti, Masha Naslidnyk, Oscar Key, Samuel Kaski, François-Xavier Briol arXiv code ICML 2023 poster
• Towards Healing the Blindness of Score Matching Mingtian Zhang, Oscar Key, Peter Hayes, David Barber, Brooks Paige, François-Xavier Briol arXiv NeurIPS 2022 workshop
• Composite Goodness-of-fit Tests with Kernels Oscar Key*, Arthur Gretton, François-Xavier Briol, Tamara Fernandez* arXiv code talk JMLR
• On Feature Collapse and Deep Kernel Learning for Single Forward Pass Joost van Amersfoort, Lewis Smith, Andrew Jesson, Oscar Key, Yarin Gal arXiv code NeurIPS 2021 workshop
• On Signal-to-Noise Ratio Issues in Variational Inference for Deep Gaussian Processes Tim G. J. Rudner*, Oscar Key*, Yarin Gal, Tom Rainforth arXiv code ICML 2021 poster
• Generating Interpretable Counterfactual Explanations By Implicit Minimisation of Epistemic and Aleatoric Uncertainties Lisa Schut*, Oscar Key*, Rory McGrath, Luca Costabello, Bogdan Sacaleanu, Medb Corcoran, Yarin Gal arXiv code AISTATS 2021 poster
• PyTorch library for constrained MPC, using the cross-entropy method code