Entities › Artificial systems

Deep reinforcement learning agents

Neural networks trained by trial and error to maximise a reward signal. They learn a policy or value function by interacting with an environment, often a simulated one such as a game.

Neural networks trained by trial and error to maximise a reward signal. They learn a policy or value function by interacting with an environment, often a simulated one such as a game. They can reach superhuman skill at narrow tasks, but they're optimised for a single reward goal.

Sources

• Mnih V, Kavukcuoglu K, Silver D, Rusu AA, et al. (2015) Human-level control through deep reinforcement learning. Nature 518(7540):529-533. doi:10.1038/nature14236.
• Silver D et al. (2016). Mastering the game of Go with deep neural networks and tree search. Nature 529:484-489. doi:10.1038/nature16961.

Assess it against the published criteria