Google DeepMind Research Scientist by field of study

This page reads the Google DeepMind Research Scientist opening in London through one lens: field of study. The role sits in the Reinforcement Learning team, but DeepMind does not hire from a single discipline — RL researchers arrive from machine learning, control theory, robotics, statistics, game theory and even neuroscience. What changes with your field is the evidence you must show and the way you should frame it. Below: the core field the role belongs to, the feeder fields that map onto it, and how applicants from adjacent disciplines position themselves.

Employer: Google DeepMind
Location: London, UK
Role Type: Full-time, Research Scientist
Deadline: Rolling / Open (apply as soon as possible)
Responsibilities: Initiate novel RL research directions, run end-to-end experiments, share knowledge with team, communicate results internally and externally, contribute to first-principles research culture
Required Qualifications: Passion for RL; research track record with peer-reviewed publications; strong coding skills; evidence of owning experiments end-to-end; PhD in ML preferred or equivalent experience
Preferred Experience: RL for sequence models, post-training, preference-based learning, or agentic systems; JAX/Flax or PyTorch; scaling experiments
Apply: job-boards.greenhouse.io/deepmind – Research Scientist, RL

Google DeepMind’s Reinforcement Learning team is a tight-knit group of scientists and engineers tackling large-scale RL research challenges. Working here means access to world-class compute, collaborators, and a mission-driven culture pushing the frontier of artificial intelligence. If you have RL expertise and a strong research track record, this is an exceptional opportunity to have real scientific and product impact at scale.

Always verify deadlines and details on the official Google DeepMind careers website before applying.

The core field: reinforcement learning

The London RL team works on the problems listed in the role itself — RL for sequence models, post-training of large models, preference-based learning and agentic systems, on top of the team’s long-running fundamental RL research. A direct-field applicant is expected to show peer-reviewed work in this space and experience owning experiments end to end, from idea to large-scale run to written result.

Feeder fields: how your discipline maps

Machine learning / AI: the direct route — deep learning research with publications at the major venues.
Control theory and robotics: dynamical systems, optimal control and real-robot learning translate directly into RL for embodied and simulated agents.
Game theory and multi-agent systems: strategic interaction and multi-agent learning are a named strength of DeepMind’s RL research.
Statistics and optimisation: bandits, sequential decision-making and stochastic optimisation are core RL mathematics.
Computational neuroscience: RL grew partly out of reward-learning research in the brain; neuroscience backgrounds with quantitative, publish-or perish research records are read seriously.
Mathematics and physics: eligible when the research record connects to learning or decision systems rather than standing alone.

Field-specific evidence the hiring team reads

Whichever field you come from, the file they read has the same shape: a research track record with peer-reviewed publications, strong coding skills, and proof that you have owned experiments end to end. What differs is calibration. A control-theory applicant is expected to connect their stability and planning results to learning systems; a neuroscientist must show the computational side of their work; a statistician should point at sequential decision problems. The preferred experience in the listing — RL for sequence models, post-training, preference-based learning, agentic systems, JAX/Flax or PyTorch at scale — is the checklist your field evidence gets matched against.

If your field is adjacent: positioning

Adjacent-field applicants strengthen their case three ways. First, translate: state your results in the vocabulary of the RL problem they solve. Second, bridge: one project, preprint or open-source artefact that visibly touches RL counts for more than a general claim of interest. Third, be specific about the team’s current directions — post-training and agentic systems are where a sequence-modelling or human-feedback background converts most directly.

Want the general version? This page covers the DeepMind Research Scientist opening from the field-of-study angle. For the complete role overview — the full team list, salary bands and the general application route — see Google DeepMind Research Scientist — the full role guide.

Frequently asked questions

Do I need a PhD in machine learning for this DeepMind role?

A PhD in ML is preferred, but the listing accepts equivalent experience. What is non-negotiable is the research record: peer-reviewed publications and experiments you owned end to end.

Which fields of study feed into DeepMind’s RL team?

Machine learning directly, plus control theory, robotics, game theory, statistics, optimisation and computational neuroscience. The team reads each field’s evidence against the RL problems the team works on.

I work in robotics and control, not RL. Can I apply?

Yes — control and robotics are recognised feeder fields. Frame your work in terms of the sequential decision problems it solves and, if you can, show one bridge into learning-based methods.

Do I need publications specifically in reinforcement learning?

Publications in RL are the strongest signal, but a strong record in a closely related area plus clear evidence you can run RL research is a credible application, especially for the team’s current post-training and agentic-systems directions.