Mehdi Mirza
Mehdi Mirza co-authored papers that span generative models and deep reinforcement learning. In 2014 he and Simon Osindero introduced conditional GANs, which feed extra data into both networks. He co-authored Generative Adversarial Nets with Ian Goodfellow and others, and NeurIPS gave that paper one of its 2024 Test of Time Paper Awards. In 2016 he co-authored Asynchronous Methods for Deep Reinforcement Learning, and ICML gave it a Test of Time Award in 2026. As of October 2026, he is a research scientist at DeepMind.
- Now at
- Google DeepMind
- Works
- 24
- Citations
- 30,603
- h-index
- 16
Publication numbers from OpenAlex, as of 15 October 2026.
Timeline
-
2014
Generative adversarial nets
He is one of eight co-authors of Generative Adversarial Nets, with Ian Goodfellow and Yoshua Bengio, a paper that NIPS publishes in 2014.
-
November 2014
Conditional GANs
In November 2014 he and Simon Osindero post Conditional Generative Adversarial Nets on arXiv, introducing conditional GANs that feed extra data into both networks.
-
2016
Asynchronous methods for deep RL
Asynchronous Methods for Deep Reinforcement Learning, written with Volodymyr Mnih and others, appears in the 2016 proceedings of the International Conference on Machine Learning.
-
September 2020
Physically embedded planning problems
In September 2020 he co-authors Physically Embedded Planning Problems, which embeds Sokoban, tic-tac-toe and Go in a physics engine.
-
2024
A NeurIPS Test of Time award
NeurIPS gives its 2024 Test of Time Paper Award to Generative Adversarial Nets, a 2014 paper that lists him among its authors.
-
2026
An ICML Test of Time award
In 2026 ICML presents its Test of Time Award for Asynchronous Methods for Deep Reinforcement Learning, a 2016 paper that lists him among its authors.
-
As of October 2026
A research scientist at DeepMind
He is a research scientist at DeepMind.
Sources and credits
Photo: We have no published photo of Mehdi Mirza yet. Send us a free-licensed photo.
- Generative Adversarial Nets
- [1411.1784] Conditional Generative Adversarial Nets
- Asynchronous Methods for Deep Reinforcement Learning
- [2009.05524] Physically Embedded Planning Problems: New Challenges for Reinforcement Learning
- Announcing the NeurIPS 2024 Test of Time Paper Awards – NeurIPS Blog
- ICML Test Of Time Asynchronous Methods for Deep Reinforcement Learning
- Mehdi Mirza | DeepAI