Isaac Reid

I'm an incoming research scientist at Google DeepMind, working on Gemini Robotics. I'm broadly interested in making transformers better, especially when learning with multimodal data and with limited access to compute. I live in London, UK.

Donostia, April 2024

PhD research. My PhD at the University of Cambridge was supervised by Dr Adrian Weller and Prof. Rich Turner, collaborating closely with Prof. Krzysztof Choromanski. It was funded by a Trinity College external studentship and a Google PhD Fellowship. I spent my time thinking about linear transformer attention for graph-structured data. To this end, I designed novel Monte Carlo estimators to approximate the softmax kernel, random walk-based algorithms for building ‘graph random features’, and new position encodings using Lie groups. I also worked on statistical physics, approximate Bayesian inference, and influence functions. I spent 6 months living in New York, helping train Gemini Robotics 1.5.

Teaching. I taught mathematics for second-year undergraduate engineers at Trinity (2023-2025), and mentored fantastic students. Since 2024, I have been a guest lecturer for the Cambridge MLMI course, providing an introduction to transformers.

Past lives. I obtained my MPhys at Hertford College, Oxford, where I submitted my dissertation on entanglement barriers in dual-unitary quantum circuits, supervised by Dr Bruno Bertini and Prof. Fabian Essler. As an undergrad, I also researched simplicity bias in deep neural networks with Prof. Ard Louis at the Rudolf Peierls Centre for Theoretical Physics, Oxford, and Bose-Einstein condensation in active matter with Dr Benoît Mahault and Prof. Ramin Golestanian at the Max Planck Institute for Dynamics and Self Organisation, Göttingen. I worked at this early-stage startup for a year, developing adaptive optics for laser fabrication of qubits inside diamonds, and advised IQ Capital as a part-time research associate.

Contact me at ir337(at)cam.ac.uk.

News

Jan 2026: Matthew Zhang's GRFs for GPs paper accepted to ICLR
Sep 2025: d-TDA accepted to NeurIPS and Gemini Robotics 1.5 released
May 2025: STRING paper accepted to ICML, and moving to NY to join GDM as a SR
Jan 2025: Two papers accepted to ICLR and one paper accepted to AISTATS
Nov 2024: Awarded a Google PhD Fellowship in Algorithms and Theory