Policy Gradient

Policy gradient methods optimize a parameterized policy π_θ directly by ascending the gradient of expected return, rather than via an action-value table.

Definition

Policy gradient methods optimize a parameterized policy π_θ directly by ascending the gradient of expected return, rather than via an action-value table.

RelatedProximal Policy Optimization Activation Function Adam Optimizer Agent Workflow Attention Mechanism Autoencoder Backpropagation Batch Normalization Beam Search BERT Byte Pair Encoding Chain of Thought CLIP Contrastive Learning

Same category · AI SystemsLong Short-Term Memory LoRA Mixture of Experts Neural Network Overfitting Perceptron Positional Encoding Prompt Injection Proximal Policy Optimization Q-Learning Recurrent Neural Network Regularization

Tagspolicy gradient REINFORCE actor-critic

More in AI Systems Wiki page Encyclopedia home

← Back to Codex