Rainbow DQN

E200562

DQN extension deep reinforcement learning algorithm value-based reinforcement learning method

Rainbow DQN is a deep reinforcement learning algorithm that combines several key extensions to the original DQN—such as double Q-learning, prioritized replay, dueling networks, multi-step learning, distributional RL, and noisy nets—into a single, more performant agent.

Try in SPARQL Jump to: Surface forms Statements Referenced by

All labels observed (3)

Label	Occurrences
Rainbow DQN canonical	3
C51 distributional DQN	1
Rainbow: Combining Improvements in Deep Reinforcement Learning	1

Statements (49)

Predicate	Object
instanceOf	DQN extension ⓘ deep reinforcement learning algorithm ⓘ value-based reinforcement learning method ⓘ
actionSpace	discrete action spaces ⓘ
basedOn	Deep Q-Learning ⓘ surface form: Deep Q-Network
citationYear	2018 ⓘ
codeAvailability	open-source implementations in multiple frameworks (e.g., PyTorch, TensorFlow) ⓘ
combines	Double Q-learning ⓘ distributional reinforcement learning ⓘ dueling network architecture ⓘ multi-step learning ⓘ noisy networks for exploration ⓘ prioritized experience replay ⓘ
developedAt	DeepMind ⓘ
evaluatedOn	Atari 2600 games from the Arcade Learning Environment ⓘ
goal	maximize expected cumulative reward ⓘ
improvesOver	Rainbow DQN self-linksurface differs ⓘ surface form: C51 distributional DQN Deep Q-Learning ⓘ surface form: DQN Double DQN ⓘ Dueling DQN ⓘ Prioritized Experience Replay DQN ⓘ surface form: Prioritized DQN
influenced	subsequent Atari benchmark baselines ⓘ
introducedInPaper	Rainbow DQN self-linksurface differs ⓘ surface form: Rainbow: Combining Improvements in Deep Reinforcement Learning
learningSignal	temporal-difference error ⓘ
optimizationAlgorithm	stochastic gradient descent variant ⓘ
outperforms	baseline DQN on Atari benchmarks ⓘ
proposedBy	Bilal Piot ⓘ Dan Horgan ⓘ David Silver ⓘ Georg Ostrovski ⓘ Hado van Hasselt ⓘ Joseph Modayil ⓘ Matteo Hessel ⓘ Mohammad Azar ⓘ Tom Schaul ⓘ Will Dabney ⓘ
publishedAt	AAAI Conference on Artificial Intelligence ⓘ surface form: AAAI 2018
taskType	model-free reinforcement learning ⓘ
uses	Q-learning update rule ⓘ epsilon-greedy policy with noisy nets modification ⓘ experience replay buffer ⓘ target network ⓘ
usesDistributionalMethod	categorical value distribution (C51) ⓘ
usesExplorationMethod	noisy linear layers ⓘ
usesFunctionApproximator	deep convolutional neural network ⓘ
usesMultiStepReturn	n-step returns ⓘ
usesNetworkArchitecture	dueling network with value and advantage streams ⓘ
usesPrioritization	proportional prioritized replay ⓘ
valueRepresentation	distribution over returns ⓘ

Referenced by (5)

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

Atari deep Q-network → inspiredAlgorithm → Rainbow DQN ⓘ

Dueling DQN → influenced → Rainbow DQN ⓘ

Double DQN → influenced → Rainbow DQN ⓘ

Rainbow DQN → introducedInPaper → Rainbow DQN self-linksurface differs ⓘ

this entity surface form: Rainbow: Combining Improvements in Deep Reinforcement Learning

Rainbow DQN → improvesOver → Rainbow DQN self-linksurface differs ⓘ

this entity surface form: C51 distributional DQN