1985–, Computer scientist
Volodymyr Mnih is a Ukrainian-Canadian computer scientist whose 2013 NIPS workshop paper and subsequent 2015 Nature paper Human-level Control through Deep Reinforcement Learning (with Kavukcuoglu, Silver, Rusu and others at DeepMind) introduced DQN, a deep convolutional network trained to play Atari 2600 games at human level using only pixels and the game score as input.
DQN was the proof-of-concept of deep reinforcement learning, combining deep neural networks with Q-learning. The two key technical innovations were experience replay (storing transitions in a buffer and sampling mini-batches to break correlations) and target networks (a slowly-updated copy of the Q-network used as the regression target). Both have been essential ingredients of deep RL ever since.
Mnih has remained at DeepMind, contributing to the ongoing programme on reinforcement learning, attention mechanisms and large-scale agents. The DQN result is among the touchstone moments of the modern AI era.
Video
Related people: David Silver, Demis Hassabis
Works cited in this book:
- Human-level control through deep reinforcement learning (2015) (with Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, Demis Hassabis)
Discussed in:
- Chapter 1: What Is AI?, A Brief History of AI