Volodymyr Mnih

Timeline

1. 2008

   A master's at Alberta

   He completes a master's degree in computing science at the University of Alberta in 2008, supervised by Csaba Szepesvári. The thesis is Efficient Stopping Rules.

2. 2010–2013

   Aerial images at Toronto

   In 2010 he and Geoffrey Hinton publish Learning to Detect Roads in High-Resolution Aerial Images at the European Conference on Computer Vision. He completes a PhD in machine learning at the University of Toronto in 2013 under Geoffrey Hinton, and the dissertation is Machine Learning for Aerial Image Labeling.

3. 2013

   Playing Atari from pixels

   Playing Atari with Deep Reinforcement Learning appears in 2013, with him, Koray Kavukcuoglu, and David Silver among the authors. The model is a convolutional neural network trained with a variant of Q-learning on raw pixels, and it surpasses a human expert on three of seven Atari 2600 games.

4. 2014

   Recurrent models of attention

   Recurrent Models of Visual Attention, which he coauthors with Nicolas Heess, Alex Graves, and Koray Kavukcuoglu, appears in 2014. The recurrent network selects a sequence of image regions and can be trained with reinforcement learning.

5. 2015

   Human-level control in Nature

   Nature publishes Human-level control through deep reinforcement learning in 2015, with him and colleagues at Google DeepMind. The deep Q-network agent, receiving only the pixels and the game score, reaches a level comparable to a professional human games tester across 49 Atari 2600 games.

6. 2016

   Asynchronous actor-critic training

   Asynchronous Methods for Deep Reinforcement Learning, which he coauthors, appears at ICML in 2016. Parallel actor-learners stabilize training, and an asynchronous variant of actor-critic surpasses the state of the art on the Atari domain while training on a single multi-core CPU.

7. 2024

   ElasticTok for video frames

   In 2024 he coauthors ElasticTok with Wilson Yan, Aleksandra Faust, Matei Zaharia, Pieter Abbeel, and Hao Liu. The method encodes a frame into a variable number of tokens, and his affiliation is Google DeepMind.

8. As of October 2026

   Research scientist at DeepMind

   He is a Canadian research scientist at Google DeepMind. In 2026 Asynchronous Methods for Deep Reinforcement Learning receives the ICML Test of Time Award.