Atari deep Q-network

E39543

deep Q-network deep reinforcement learning algorithm model-free reinforcement learning method off-policy reinforcement learning method value-based reinforcement learning method

The Atari deep Q-network is a pioneering deep reinforcement learning system that learned to play a wide range of Atari 2600 video games directly from raw pixels at human-level or better performance.

Jump to: Surface forms Statements Referenced by

Observed surface forms (3)

Surface form	Occurrences
DQN	3
Deep Q-Network	2
Playing Atari with Deep Reinforcement Learning	1

Statements (50)

Predicate	Object
instanceOf	deep Q-network ⓘ deep reinforcement learning algorithm ⓘ model-free reinforcement learning method ⓘ off-policy reinforcement learning method ⓘ value-based reinforcement learning method ⓘ
achievedPerformanceLevel	human-level control on many Atari 2600 games ⓘ
actionSpace	discrete actions ⓘ
basedOnAlgorithm	Q-learning ⓘ
coAuthor	Alex Graves ⓘ Daan Wierstra ⓘ David Silver ⓘ Ioannis Antonoglou ⓘ Koray Kavukcuoglu ⓘ Martin Riedmiller ⓘ
developedBy	DeepMind ⓘ
doesNotUse	game-specific prior knowledge ⓘ hand-crafted features ⓘ
domain	Atari 2600 video games ⓘ
environmentFramework	Arcade Learning Environment ⓘ
evaluationMetric	average score over episodes ⓘ
evaluationSetting	same network architecture across games ⓘ single set of hyperparameters across games ⓘ
firstAuthor	Volodymyr Mnih ⓘ
inputFrameSize	84x84 grayscale images ⓘ
inputFrameStack	4 consecutive frames ⓘ
inputSource	Atari 2600 games ⓘ
inputType	raw pixel images ⓘ
inspiredAlgorithm	Double DQN ⓘ Dueling DQN ⓘ Prioritized Experience Replay DQN ⓘ surface form: Prioritized Experience Replay Rainbow DQN ⓘ
introducedInPaper	Atari deep Q-network self-linksurface differs ⓘ surface form: Playing Atari with Deep Reinforcement Learning
introducedInYear	2013 ⓘ
journalPublicationYear	2015 ⓘ
learningParadigm	trial-and-error learning ⓘ
notableContribution	demonstrated deep learning can learn control policies directly from high-dimensional sensory input ⓘ introduced target networks for stabilizing deep Q-learning ⓘ popularized experience replay in deep reinforcement learning ⓘ
observationType	screen images only ⓘ
outputType	Q-values for discrete actions ⓘ action-value function ⓘ
publishedInJournal	Nature ⓘ
rewardSignal	game score changes ⓘ
trainingSignal	game score reward ⓘ
usesExplorationStrategy	epsilon-greedy policy ⓘ
usesFunctionApproximator	convolutional neural network ⓘ
usesLossFunction	temporal-difference error ⓘ
usesOptimizationMethod	stochastic gradient descent ⓘ
usesStabilizationTechnique	experience replay ⓘ target network ⓘ

Referenced by (7)

Full triples — surface form annotated when it differs from this entity's canonical label.

A3C → comparedWith → Atari deep Q-network ⓘ

this entity surface form: DQN

DeepMind → developed → Atari deep Q-network ⓘ

this entity surface form: Deep Q-Network

OpenAI Baselines → implementsAlgorithm → Atari deep Q-network ⓘ

this entity surface form: Deep Q-Network

OpenAI Baselines → implementsAlgorithm → Atari deep Q-network ⓘ

this entity surface form: DQN

Atari deep Q-network → introducedInPaper → Atari deep Q-network self-linksurface differs ⓘ

this entity surface form: Playing Atari with Deep Reinforcement Learning

DeepMind → knownFor → Atari deep Q-network ⓘ

TF-Agents → supportsAlgorithmFamily → Atari deep Q-network ⓘ

this entity surface form: DQN