PeopleofSuperintelligence

Researcher

Matteo Hessel

Matteo Hessel is a reinforcement learning researcher at Google DeepMind whose papers cover value-based agents, multi-task agents and the automatic discovery of learning rules. He coauthored the dueling-network paper that was a best paper at ICML 2016, and he was first author of Rainbow in 2017, which combined six DQN extensions in one Atari agent. His 2018 PopArt paper trained a single policy across 57 Atari games, and his 2021 Muesli paper matched MuZero's Atari results without deep search. In 2020 he coauthored the Learned Policy Gradient work on discovering RL algorithms, and in 2025 he was an author of the Gemini 2.5 report and of the Nature paper on discovering state-of-the-art reinforcement learning algorithms. He works as a research engineer at Google DeepMind.

Timeline

  1. 2016

    A best paper on dueling networks

    He coauthors 'Dueling Network Architectures for Deep Reinforcement Learning' with Ziyu Wang, Tom Schaul, Hado van Hasselt, Marc Lanctot and Nando de Freitas, and the paper is a best paper at ICML 2016. Its network keeps one estimator for the state value function and one for the state-dependent action advantage function, and its agent outperforms the state of the art on Atari 2600.

  2. 2017

    The Rainbow agent

    In October 2017, he is first author of 'Rainbow: Combining Improvements in Deep Reinforcement Learning', which examines six extensions to the DQN algorithm and studies their combination. The combination gives state-of-the-art performance on the Atari 2600 benchmark in both data efficiency and final performance.

  3. 2018

    PopArt for multi-task Atari

    In September 2018, he is first author of 'Multi-task Deep Reinforcement Learning with PopArt', which adapts each task's contribution to the agent's updates so that all tasks have a similar impact on learning. A single trained policy with one set of weights exceeds median human performance on a set of 57 Atari games.

  4. March 2019

    A seminar at Politecnico di Milano

    In March 2019, he gives the talk 'Deep Reinforcement Learning and Artificial Intelligence' at Politecnico di Milano as a Research Engineer at DeepMind. His research at DeepMind focuses on building general artificial agents, and his past work includes Dueling Networks, PopArt, Predictron and Rainbow.

  5. 2020

    The Learned Policy Gradient

    In 2020, he coauthors 'Discovering Reinforcement Learning Algorithms' with Junhyuk Oh, Wojciech Czarnecki, Zhongwen Xu, Hado van Hasselt, Satinder Singh and David Silver. The meta-learning method discovers an entire update rule, covering what to predict and how to learn from it, and the resulting Learned Policy Gradient algorithm generalises from toy environments to Atari games.

  6. 2021

    The Muesli policy update

    In April 2021, he is first author of 'Muesli: Combining Improvements in Policy Optimization', written with David Silver, Hado van Hasselt and others. The Muesli update combines regularized policy optimization with model learning as an auxiliary loss, and it matches MuZero's state-of-the-art performance on Atari without deep search.

  7. 2025

    Gemini 2.5 and a Nature paper

    In July 2025, he is among the authors of the Gemini 2.5 report, which introduces Gemini 2.5 Pro and Gemini 2.5 Flash. In October 2025, Nature publishes 'Discovering state-of-the-art reinforcement learning algorithms', coauthored with colleagues at Google DeepMind, which shows that a discovered RL rule can outperform manually designed rules.

  8. As of October 2026

    Research engineer at Google DeepMind

    As of 2026, he works as a research engineer at Google DeepMind.

  1. 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.
  2. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. 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.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Google. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Awards | ICML New York City icml.cc. Checked 21 October 2026.
  2. [1511.06581] Dueling Network Architectures for Deep Reinforcement Learning arxiv.org. Checked 21 October 2026.
  3. [1710.02298] Rainbow: Combining Improvements in Deep Reinforcement Learning arxiv.org. Checked 21 October 2026.
  4. [1809.04474] Multi-task Deep Reinforcement Learning with PopArt arxiv.org. Checked 21 October 2026.
  5. Deep Reinforcement Learning and Artificial Intelligence - Events - DEIB deib.polimi.it. Checked 21 October 2026.
  6. [2007.08794] Discovering Reinforcement Learning Algorithms arxiv.org. Checked 21 October 2026.
  7. [2104.06159] Muesli: Combining Improvements in Policy Optimization arxiv.org. Checked 21 October 2026.
  8. [2507.06261] Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities arxiv.org. Checked 21 October 2026.
  9. Discovering state-of-the-art reinforcement learning algorithms | Nature nature.com. Checked 21 October 2026.
  10. Matteo Hessel - Google Scholar scholar.google.com. Checked 21 October 2026.