Trust Region Policy Optimization
GPTKB entity
Statements (49)
| Predicate | Object |
|---|---|
| gptkbp:instanceOf |
algorithm
|
| gptkbp:appliesTo |
reinforcement learning
|
| gptkbp:developedBy |
gptkb:John_Schulman
|
| https://www.w3.org/2000/01/rdf-schema#label |
Trust Region Policy Optimization
|
| gptkbp:improves |
policy gradient methods
|
| gptkbp:isAvenueFor |
continuous action spaces
discrete action spaces |
| gptkbp:isChallengedBy |
high-dimensional spaces
local optima non-stationary environments |
| gptkbp:isEvaluatedBy |
real-world applications
stability simulated environments convergence rate sample efficiency robotic tasks Atari_games |
| gptkbp:isExploredIn |
academic research
industry applications AI conferences machine learning workshops reinforcement learning symposiums |
| gptkbp:isInfluencedBy |
trust region methods
natural gradient methods |
| gptkbp:isLocatedIn |
gptkb:PyTorch
TensorFlow |
| gptkbp:isPartOf |
gptkb:REINFORCE
gptkb:DDPG policy optimization family TRPO_variants |
| gptkbp:isRelatedTo |
actor-critic methods
policy optimization |
| gptkbp:isSupportedBy |
empirical results
theoretical guarantees |
| gptkbp:isUsedFor |
improving performance
exploration strategies training agents |
| gptkbp:isUsedIn |
robotics
game playing |
| gptkbp:isVisitedBy |
robust learning
sample-efficient learning scalable learning |
| gptkbp:maxRange |
expected reward
|
| gptkbp:performance |
policy parameters
|
| gptkbp:publishedIn |
2015
|
| gptkbp:reduces |
KL divergence
|
| gptkbp:relatedTo |
Proximal Policy Optimization
|
| gptkbp:requires |
second-order information
|
| gptkbp:uses |
trust region methods
|