PeopleofSuperintelligence

Researcher

Simon Schmitt

Simon Schmitt is a reinforcement-learning researcher whose work spans agents that learn across tasks, plan with learned models and estimate uncertainty. He coauthored teacher-guided learning and PopArt in 2018, MuZero in 2019–2020, and Muesli and Sampled MuZero in 2021. His later work studied efficient exploration and neural-network uncertainty. A Nature paper reached its version of record in March 2026, listing his affiliation as Google DeepMind in London and naming him as a coauthor of AlphaProof, an agent that learns formal proofs through reinforcement learning.

Works
30
Citations
1,955
h-index
13
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2018

    Kickstarting deep reinforcement learning

    In 2018, he coauthors Kickstarting Deep Reinforcement Learning, which trains a student with guidance from previously trained teachers. On DMLab-30, the method improves data efficiency and lets one student surpass a from-scratch agent’s final performance by 42 percent.

  2. 2018

    PopArt for Atari

    In 2018, he coauthors the PopArt work, which adapts value-target normalization to differences in reward scales across games. Applied to IMPALA, it enables one agent to exceed human median performance across 57 Atari games.

  3. 2019–2020

    MuZero learns to plan

    In 2019, he coauthors MuZero, which combines tree-based search with a learned model to plan in unknown environments. In 2020, the system reaches a new state of the art on Atari and matches AlphaZero’s performance in Go, chess and shogi without game rules.

  4. 2021

    Muesli matches Atari performance

    At ICML 2021, he coauthors Muesli, a policy update that matches MuZero’s Atari performance without deep search and uses computation at a speed comparable to model-free baselines.

  5. 2021

    Sampled MuZero for complex actions

    In 2021, he coauthors Sampled MuZero, which plans over sampled actions to learn in environments with arbitrarily complex action spaces.

  6. 2023

    Epistemic value estimation

    In 2023, he coauthors EVE, a method that gives reinforcement-learning agents a tractable posterior for estimating value uncertainty. Experiments show that EVE supports efficient exploration in hard-exploration tasks.

  7. 2025

    Delta Variances for uncertainty

    In 2025, he coauthors Delta Variances, a computationally efficient approach to estimating epistemic uncertainty in neural networks. It produces competitive results in a weather simulator using one gradient computation.

  8. March 2026

    Formal proofs with AlphaProof

    The paper reaches its version of record in March 2026, with his affiliation listed as Google DeepMind in London and his name among its coauthors. AlphaProof learns formal proofs through reinforcement learning; the paper first appears online in November 2025.

Career

  1. 2022–2025University College London
  2. 2018–2025Google DeepMind
  3. 2012–2017Freie Universität Berlin

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

  1. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Simon Schmitt. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  2. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Simon Schmitt. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  3. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Simon Schmitt. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  4. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Simon Schmitt yet. Send us a free-licensed photo.

  1. Kickstarting Deep Reinforcement Learning arxiv.org. Checked 22 October 2026.
  2. Preserving Outputs Precisely while Adaptively Rescaling Targets — Google DeepMind deepmind.google. Checked 22 October 2026.
  3. Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model arxiv.org. Checked 22 October 2026.
  4. MuZero: Mastering Go, chess, shogi and Atari without rules — Google DeepMind deepmind.google. Checked 22 October 2026.
  5. Muesli: Combining Improvements in Policy Optimization proceedings.mlr.press. Checked 22 October 2026.
  6. Learning and Planning in Complex Action Spaces proceedings.mlr.press. Checked 22 October 2026.
  7. Exploration via Epistemic Value Estimation | Proceedings of the AAAI Conference on Artificial Intelligence ojs.aaai.org. Checked 22 October 2026.
  8. Delta Variances — Google DeepMind deepmind.google. Checked 22 October 2026.
  9. General Uncertainty Estimation with Delta Variances arxiv.org. Checked 22 October 2026.
  10. Olympiad-level formal mathematical reasoning with reinforcement learning | Nature nature.com. Checked 22 October 2026.