Decision-making and artificial general intelligence using deep neural networks

Deep neural networks have been remarkably successful models that have led to breakthroughs in a wide variety tasks, including classification, density estimation and reinforcement learning. Various reinforcement learning algorithms have used deep neural networks to master a variety of computer games, ranging from simple Atari games to complex 3D navigation tasks. Progress in deep learning has been driven by improved neural network architectures and training methods using insights from linear algebra, probability theory, statistics, and optimization theory. For example, the combination of Monte Carlo tree search and deep learning allowed AlphaGo to master the game of Go. In this talk I will give a broad overview of our research at DeepMind, including both the areas of deep and reinforcement learning.

Industrial Seminar

• Speaker: Audrunas Gruslys, Google Deep Mind
• Tuesday 07 February 2017, 16:00–17:00
• Venue: MR14.
• Series: Cambridge Centre for Analysis talks; organiser: CCA.