Maximum Entropy Bayesian Actor Critic

Steven Thomas Homer

Maximum Entropy Bayesian Actor Critic

Onderzoeksoutput: Poster

26 Downloads (Pure)

Samenvatting

Maximum Entropy Bayesian Actor Critic aims to avoid the problems of sample inefficiency and convergence brittleness by combining Bayesian and Maximum Entropy Reinforcement Learning. A Soft Policy Gradient is derived from a soft value function that introduces an entropy-weighted term to the policy gradient theorem. A Max Entropy Parameter Update alters the Bayesian actor critic parameter update rule to account for the additional policy entropy term.

Originele taal-2	English
Status	Published - 7 nov 2019
Evenement	BNAIC 2019 - Brussels, Belgium Duur: 7 nov 2019 → 8 nov 2019

Conference

Conference	BNAIC 2019
Land/Regio	Belgium
Stad	Brussels
Periode	7/11/19 → 8/11/19

Toegang tot document

Maximum Entropy Bayesian Actor CriticFinal published version, 164 KB

1 Conference paper

Maximum Entropy Bayesian Actor Critic
Homer, S. T., 6 nov 2019, BNAIC/Benelearn 2019. Vol. 2491. 12 blz. (CEUR Workshop Proceedings).
Onderzoeksoutput: Conference paper › Research

Citeer dit

@conference{d26779d9990b452896c61a40c5482129,

title = "Maximum Entropy Bayesian Actor Critic",

abstract = "Maximum Entropy Bayesian Actor Critic aims to avoid the problems of sample inefficiency and convergence brittleness by combining Bayesian and Maximum Entropy Reinforcement Learning. A Soft Policy Gradient is derived from a soft value function that introduces an entropy-weighted term to the policy gradient theorem. A Max Entropy Parameter Update alters the Bayesian actor critic parameter update rule to account for the additional policy entropy term.",

keywords = "Reinforcement Learning, Bayesian Inference, Maximum Entropy",

author = "Homer, {Steven Thomas}",

year = "2019",

month = nov,

day = "7",

language = "English",

note = "BNAIC 2019 ; Conference date: 07-11-2019 Through 08-11-2019",

}

TY - CONF

T1 - Maximum Entropy Bayesian Actor Critic

AU - Homer, Steven Thomas

PY - 2019/11/7

Y1 - 2019/11/7

N2 - Maximum Entropy Bayesian Actor Critic aims to avoid the problems of sample inefficiency and convergence brittleness by combining Bayesian and Maximum Entropy Reinforcement Learning. A Soft Policy Gradient is derived from a soft value function that introduces an entropy-weighted term to the policy gradient theorem. A Max Entropy Parameter Update alters the Bayesian actor critic parameter update rule to account for the additional policy entropy term.

AB - Maximum Entropy Bayesian Actor Critic aims to avoid the problems of sample inefficiency and convergence brittleness by combining Bayesian and Maximum Entropy Reinforcement Learning. A Soft Policy Gradient is derived from a soft value function that introduces an entropy-weighted term to the policy gradient theorem. A Max Entropy Parameter Update alters the Bayesian actor critic parameter update rule to account for the additional policy entropy term.

KW - Reinforcement Learning

KW - Bayesian Inference

KW - Maximum Entropy

M3 - Poster

T2 - BNAIC 2019

Y2 - 7 November 2019 through 8 November 2019

ER -

Maximum Entropy Bayesian Actor Critic

Samenvatting

Conference

Toegang tot document

Vingerafdruk

Onderzoekersoutput

Maximum Entropy Bayesian Actor Critic

Citeer dit