A2C

E98476

actor-critic method policy gradient method reinforcement learning algorithm

A2C (Advantage Actor-Critic) is a popular synchronous policy gradient reinforcement learning algorithm that combines value-based and policy-based methods to improve training stability and efficiency.

Try in SPARQL Jump to: Surface forms Statements Referenced by

All labels observed (1)

Label	Occurrences
A2C canonical	4

Statements (48)

Predicate	Object
instanceOf	actor-critic method ⓘ policy gradient method ⓘ reinforcement learning algorithm ⓘ
actorOutputs	action probabilities ⓘ
actorUpdatedWith	advantage-weighted log-probabilities ⓘ
advantageDefinition	A(s,a) = Q(s,a) - V(s) ⓘ
canHandle	continuous observation spaces ⓘ discrete action spaces ⓘ high-dimensional state spaces ⓘ
canUse	multiple parallel environments ⓘ
category	deep reinforcement learning ⓘ
combines	policy-based methods ⓘ value-based methods ⓘ
criticOutputs	state-value estimate ⓘ
criticTrainedWith	regression to returns or bootstrapped targets ⓘ
entropyBonusPurpose	encourage exploration ⓘ
fullName	Asynchronous Advantage Actor-Critic ⓘ surface form: Advantage Actor-Critic
goal	improve sample efficiency ⓘ improve training stability ⓘ reduce gradient variance ⓘ
implementedIn	OpenAI Baselines ⓘ PyTorch-based RL libraries ⓘ Stable Baselines ⓘ Stable Baselines ⓘ surface form: Stable Baselines3 TensorFlow-based RL libraries ⓘ
isOnPolicy	true ⓘ
isPolicyBased	true ⓘ
isRelatedTo	A3C ⓘ
isSynchronous	true ⓘ
isSynchronousVariantOf	A3C ⓘ
isValueBased	true ⓘ
optimizes	stochastic policy ⓘ
reducesVarianceUsing	advantage estimation ⓘ value function baseline ⓘ
trainingSignal	temporal-difference error ⓘ
typicalUseCase	Atari game playing ⓘ continuous control tasks ⓘ discrete action tasks ⓘ
updateFrequency	multiple environment steps per update ⓘ
usesAdvantageFunction	true ⓘ
usesBaseline	state-value function ⓘ
usesFunctionApproximator	neural network ⓘ
usesLearningParadigm	model-free reinforcement learning ⓘ
usesLossComponent	entropy regularization ⓘ policy loss ⓘ value loss ⓘ
usesObjective	policy gradient objective ⓘ
usesUpdateType	synchronous gradient updates ⓘ

Referenced by (4)

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

OpenAI Baselines → implementsAlgorithm → A2C ⓘ

Stable Baselines → supportsAlgorithm → A2C ⓘ

PPO → relatedTo → A2C ⓘ

Generalized Advantage Estimation → compatibleWith → A2C ⓘ