Volodymyr Mnih

Career

Mnih completed a master's degree in computing science at the University of Alberta, supervised by Csaba Szepesvari, with a 2008 thesis on efficient stopping rules. Mnih then earned a PhD in machine learning at the University of Toronto under Geoffrey Hinton, with a 2013 thesis on machine learning for aerial image labeling, and is now a research scientist at Google DeepMind.¹

Research

Mnih's doctoral work with Hinton applied deep learning to aerial imagery, including detecting roads in high-resolution aerial images (ECCV 2010) and learning to label aerial images from noisy data (ICML 2012), with datasets released from the homepage. Mnih also wrote CUDAMat, a CUDA-based matrix class for Python, described in a 2009 technical report.¹

Mnih was first author of a 2013 NIPS Deep Learning Workshop paper in which a convolutional network, trained with a variant of Q-learning, learned to play Atari 2600 games from raw pixels, and of the 2015 Nature paper Human-level Control through Deep Reinforcement Learning. Mnih also first-authored Recurrent Models of Visual Attention (NeurIPS 2014) and the 2016 paper Asynchronous Methods for Deep Reinforcement Learning, which presented asynchronous variants of standard reinforcement-learning algorithms, including actor-critic.¹²⁴³

Work and contributions

• Deep Q-network (DQN) Published research

  First author of the 2013 paper "Playing Atari with Deep Reinforcement Learning" and of the 2015 Nature paper "Human-level Control through Deep Reinforcement Learning", for which the homepage links code.²¹

• Asynchronous Methods for Deep Reinforcement Learning (A3C) Published research

  First author of the 2016 paper presenting asynchronous variants of four reinforcement learning algorithms trained with parallel actor-learners.³

• Recurrent Models of Visual Attention Published research

  First author, with Nicolas Heess, Alex Graves and Koray Kavukcuoglu, of the NeurIPS 2014 paper on a recurrent model that attends to a sequence of image regions.⁴¹

Timeline

1. 2008Completed a master's thesis on efficient stopping rules at the University of Alberta.¹
2. 2010Presented Learning to Detect Roads in High-Resolution Aerial Images with Geoffrey Hinton at ECCV 2010.¹
3. 2013Completed a PhD thesis at the University of Toronto on machine learning for aerial image labeling.¹
4. Dec 19, 2013Posted Playing Atari with Deep Reinforcement Learning as first author.²
5. Jun 24, 2014Posted Recurrent Models of Visual Attention, published at NeurIPS 2014.⁴¹
6. 2015Published Human-level Control through Deep Reinforcement Learning in Nature.¹
7. Feb 4, 2016Posted Asynchronous Methods for Deep Reinforcement Learning as first author.³

Where they work

• Google DeepMind 64 people in the wiki

Related people

• Adam GleaveAlso at Google DeepMind
• Timothy LillicrapAlso at Google DeepMind
• Oriol VinyalsAlso at Google DeepMind
• Tuomas HaarnojaAlso at Google DeepMind
• Richard SuttonAlso at Google DeepMind
• Alexander WettigReinforcement learning
• Aravind SrinivasReinforcement learning
• Eric MitchellReinforcement learning
• Weizhu ChenReinforcement learning
• Katja HofmannReinforcement learning