Bio

Daan Wierstra is a Dutch computer scientist and senior researcher in deep reinforcement learning, evolutionary methods for neural networks, and generative modeling. He was the first research scientist hired at DeepMind in 2011 and a senior figure across more than a decade of the company's research output, with co-authorship credits including the DDPG continuous-control paper and contributions to the variational autoencoder line. He was a co-founder of H, the Paris-headquartered agentic-AI company he established in 2023 with Charles Kantor, Laurent Sifre, Karl Tuyls, and Julien Perolat, before departing in late 2024 alongside two of the other co-founders. As of May 2026, his published record reflects more than two decades of contributions to neural-network research at IDSIA and DeepMind.

At a glance

• Education: Bachelor of computer science, Utrecht University (2004); PhD in informatics, Università della Svizzera italiana at the Dalle Molle Institute for Artificial Intelligence (IDSIA) under Jürgen Schmidhuber.
• Past roles: First research scientist at DeepMind, 2011 onward; Principal Scientist at DeepMind through 2023; Co-founder, H, 2023 to late 2024.
• Current role: Per public reporting, returned to senior research roles after the late-2024 H departure.
• Key contributions: Natural Evolution Strategies (with Tom Schaul, Tobias Glasmachers, Yi Sun, and Schmidhuber); co-author of Continuous control with deep reinforcement learning (DDPG, 2015); co-author of Stochastic Backpropagation and Approximate Inference in Deep Generative Models (Rezende, Mohamed, Wierstra, 2014); contributions across the DeepMind research output spanning relational reasoning, world models, and reinforcement-learning foundations.
• Google Scholar: Daan Wierstra
• DBLP: Daan Wierstra

Origins

Wierstra completed a bachelor's degree in computer science at Utrecht University in the Netherlands in 2004 before moving to the Dalle Molle Institute for Artificial Intelligence (IDSIA) in Lugano, Switzerland for doctoral work under Jürgen Schmidhuber. The IDSIA group, run by Schmidhuber alongside Faustino Gomez, Tom Schaul, and a senior cohort of researchers in evolutionary methods and recurrent-neural-network training, produced a substantial proportion of the senior deep-learning research diaspora that staffed DeepMind in its founding period.

The doctoral research program addressed evolutionary methods for neural-network optimization. The principal artifact from the IDSIA period is Natural Evolution Strategies, the family of black-box optimization algorithms that use the natural gradient to update a parameterized search distribution toward higher expected fitness. The technique was developed with Tom Schaul, Tobias Glasmachers, Yi Sun, and Schmidhuber and published in the Journal of Machine Learning Research in 2014.

Career

Wierstra was the first research scientist hired at DeepMind in 2011, joining the company's earliest research staff alongside the co-founders Demis Hassabis, Shane Legg, and Mustafa Suleyman. The DeepMind group at that point had not yet shipped the Atari deep-Q-network results that would establish the company's research profile, and the early research program built on the IDSIA-and-Schmidhuber-school recurrent-neural-network and evolutionary-methods lineage that several DeepMind early hires shared with Alex Graves and Tom Schaul.

The DeepMind tenure produced a substantial research record. The 2014 paper Stochastic Backpropagation and Approximate Inference in Deep Generative Models by Danilo Rezende, Shakir Mohamed, and Wierstra introduced one of the early formulations of variational inference for deep generative models, parallel to the Auto-Encoding Variational Bayes paper by Diederik Kingma and Max Welling at the University of Amsterdam. The two papers together formalized the variational-autoencoder family that became foundational across deep-generative-modeling research.

The 2015 paper Continuous control with deep reinforcement learning, with Timothy Lillicrap, Jonathan Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, and Wierstra as co-authors, introduced the Deep Deterministic Policy Gradient (DDPG) algorithm. DDPG is an actor-critic, model-free method based on the deterministic policy gradient that operates over continuous action spaces; the same paper described robust solutions across more than 20 simulated physics tasks. The DDPG line became a foundational reference for continuous-control reinforcement learning across the late 2010s.

Subsequent contributions in the DeepMind period spanned relational recurrent neural networks, imagination-augmented agents, world models, scheduling problems, embodied agents, and scalability in simulated environments. Wierstra rose to Principal Scientist during the tenure.

In 2023 Wierstra left DeepMind to co-found H in Paris with Charles Kantor, Laurent Sifre, Karl Tuyls, and Julien Perolat. The May 2024 $220 million seed round, led by Accel with Amazon, UiPath, Bpifrance, Innovation Endeavors, FirstMark, and Bernard Arnault participating, was the largest AI seed in European history at the time. The five-co-founder configuration combined Kantor's academic-and-finance background with four senior DeepMind veterans.

In late 2024, Wierstra, Tuyls, and Perolat departed H over what the company described as "operational differences." Per Sifted and other industry coverage, the three founders left H within months of each other in the second half of 2024, leaving Sifre and Kantor as the remaining co-founders. The departures were reported as a structurally significant first-year governance event for the company. Per public reporting, Wierstra returned to senior research roles after the H departure.

Affiliations

• Utrecht University: Bachelor in computer science, awarded 2004.
• Dalle Molle Institute for Artificial Intelligence (IDSIA) and Università della Svizzera italiana: PhD candidate in informatics under Jürgen Schmidhuber.
• DeepMind: First research scientist (2011); Principal Scientist through 2023.
• H: Co-founder, 2023 to late 2024.

Notable contributions

• Natural Evolution Strategies (Journal of Machine Learning Research, 2014). Co-authored with Tom Schaul, Tobias Glasmachers, Yi Sun, and Jürgen Schmidhuber, the paper introduced the family of black-box optimization algorithms using the natural gradient on parameterized search distributions.
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models (ICML 2014). Co-authored with Danilo Rezende and Shakir Mohamed, the paper introduced one of the principal early formulations of variational inference for deep generative models. The variational-autoencoder line built on this and the parallel Kingma-Welling paper became foundational across deep generative modeling.
• Continuous control with deep reinforcement learning (DDPG, ICLR 2016). Co-authored with Lillicrap, Hunt, Pritzel, Heess, Erez, Tassa, and Silver, the paper introduced the DDPG continuous-control algorithm, which became a foundational reference for actor-critic continuous-control reinforcement learning.
• DeepMind early-tenure research output. First research scientist at DeepMind in 2011, contributing across recurrent neural networks, world models, relational reasoning, imagination-augmented agents, and reinforcement-learning foundations.
• H co-founding (2023). Co-founded the Paris-headquartered agentic-AI company.

Position in the field

As of May 2026, Wierstra occupies a distinctive position as a senior deep-learning researcher with co-authorship credit on three of the structurally foundational research artifacts of the contemporary deep-learning era (Natural Evolution Strategies, the variational-autoencoder line, and DDPG) and a more than decade-long DeepMind tenure that placed him at the company from its earliest research staffing. The IDSIA-school lineage through Schmidhuber, alongside fellow alumni Alex Graves and Tom Schaul, ties him to one of the principal academic origin nodes of contemporary deep learning.

The H founding tenure and subsequent departure place Wierstra inside the post-DeepMind senior-research diaspora that has staffed several European AI insurgents through the 2023 to 2026 period. Industry coverage has characterized the H departures as a structurally significant first-year governance event, with the subsequent trajectories of Wierstra, Tuyls, and Perolat as a watchable signal for the senior research-talent dynamics of the European AI ecosystem.

Outlook

Open questions over the next 6 to 18 months:

• Future research role. Whether Wierstra returns to a senior research role at a frontier lab, an academic position, or another insurgent venture.
• Continued publication record. Continued co-authored research output following the H departure.
• Network with H peers. The trajectory of the broader Wierstra-Tuyls-Perolat post-departure cohort and any joint subsequent ventures.
• DeepMind alumni network dynamics. The senior research-talent dynamics of the post-DeepMind diaspora through 2026 and 2027.

Sources

• Daan Wierstra on Google Scholar. Publication record and citation index.
• Daan Wierstra on DBLP. Computer Science publication index.
• Daan Wierstra on Research.com. Computer Science research index covering H-index, publications, and awards.
• Natural Evolution Strategies. The 2014 JMLR paper introducing NES, co-authored with Schaul, Glasmachers, Sun, and Schmidhuber.
• Continuous control with deep reinforcement learning. The 2015 paper introducing DDPG, with Wierstra as a co-author.
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models. The 2014 paper co-authored with Rezende and Mohamed introducing the variational-inference framework for deep generative models.
• Google acquires artificial intelligence start-up for 500 million dollars. Università della Svizzera italiana feed item noting Wierstra as one of the IDSIA alumni among DeepMind's founding research staff.
• Sifted: Three cofounders leave French AI startup H. Coverage of the late-2024 H Company co-founder departures.
• TechCrunch: French AI startup H raises $220M seed round. Seed-round announcement covering the founding team.
• H (company) on Wikipedia. Company reference covering co-founder departures.