PeopleofSuperintelligence

Researcher

Adria Puigdomenech Badia

Also known as Adrià Puigdomènech

Adria Puigdomenech Badia is an AI researcher whose work spans deep reinforcement learning and multimodal models. He coauthored asynchronous reinforcement-learning research, Neural Episodic Control and Imagination-Augmented Agents; Agent57 scored above the human baseline on all 57 Atari games. His 2020 Never Give Up paper doubled a base agent's performance on hard-exploration Atari games. In 2022 he coauthored Human-level Atari 200x faster; he also coauthored Gemini in 2023 and Gemini 1.5 in 2024. ICML gave his 2016 paper the 2026 Test of Time Award, citing its influence on RL post-training for large language models.

Timeline

  1. 2016

    Asynchronous deep reinforcement learning

    In 2016, he coauthors Asynchronous Methods for Deep Reinforcement Learning, which presents asynchronous versions of four standard reinforcement-learning algorithms. Its actor-critic variant surpasses Atari's current state of the art while training for half the time on a single multi-core CPU.

  2. 2017

    Neural Episodic Control

    Neural Episodic Control, which he coauthors in 2017, is a deep reinforcement-learning agent that rapidly assimilates new experiences. Its semi-tabular value representation pairs a buffer of past experience with rapidly updated estimates, helping it learn quickly across many environments.

  3. 2017

    Imagination-Augmented Agents

    He coauthors Imagination-Augmented Agents for Deep Reinforcement Learning in 2017, an architecture that combines model-free and model-based methods. The agents use predictions from a learned environment model to build implicit plans.

  4. 2020

    Agent57 on Atari

    In 2020, he coauthors Agent57, the first deep reinforcement-learning agent to score above the human baseline on all 57 Atari 2600 games. Its meta-controller adapts exploration and long- and short-term behavior across a family of policies.

  5. 2020

    Never Give Up exploration

    He coauthors Never Give Up: Learning Directed Exploration Strategies in 2020. Its method doubles the base agent's performance on hard-exploration games across the Atari-57 suite while maintaining a high score on the other games.

  6. 2022

    Human-level Atari with less data

    Taking Agent57 as its starting point, the 2022 paper Human-level Atari 200x faster, which he coauthors, reduces by 200-fold the experience needed to exceed the human baseline.

  7. 2023–2024

    Gemini and Gemini 1.5

    He coauthors Gemini, a family of multimodal models, in 2023. In 2024, he coauthors Gemini 1.5, whose report describes reasoning over millions of tokens and retrieval above 99% up to at least 10 million tokens.

  8. 2026

    A Test of Time Award

    In 2026, ICML gives Asynchronous Methods for Deep Reinforcement Learning its Test of Time Award. The award cites the paper's parallel actor-learner method as a major contributor to reinforcement learning in LLM post-training and says it reshaped RL practice.

  1. Google had the most people shaping AI every year from 2014 to 2025Our 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. The United States employs an estimated 48% of people shaping AIWe estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.
  3. 43% of living people shaping AI work outside their country of birthWe estimate that 43% of living people shaping AI work outside their country of birth (520 of 1,198 people with known birth and employment countries).
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionWe estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.

Sources and credits

Photo: We have no published photo of Adria Puigdomenech Badia yet. Send us a free-licensed photo.

  1. [1602.01783] Asynchronous Methods for Deep Reinforcement Learning arxiv.org. Checked 21 October 2026.
  2. [1703.01988] Neural Episodic Control arxiv.org. Checked 21 October 2026.
  3. [1707.06203] Imagination-Augmented Agents for Deep Reinforcement Learning arxiv.org. Checked 21 October 2026.
  4. Agent57: Outperforming the human Atari benchmark — Google DeepMind deepmind.google. Checked 21 October 2026.
  5. [2002.06038] Never Give Up: Learning Directed Exploration Strategies arxiv.org. Checked 21 October 2026.
  6. [2209.07550] Human-level Atari 200x faster arxiv.org. Checked 21 October 2026.
  7. [2312.11805] Gemini: A Family of Highly Capable Multimodal Models arxiv.org. Checked 21 October 2026.
  8. [2403.05530] Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context arxiv.org. Checked 21 October 2026.
  9. Announcing the ICML 2026 Awards – ICML Blog blog.icml.cc. Checked 21 October 2026.