AI 용어집
인공지능 완전 사전
Hierarchical Actor-Critic (HAC)
Reinforcement learning architecture combining multi-level hierarchical actors and critics to solve complex tasks through temporal decomposition.
High-level Policy
Decision policy at the top of the hierarchy that selects subgoals or options to guide lower-level policies.
Low-level Policy
Base policy in the hierarchy that executes primitive actions to achieve the subgoals defined by the higher-level policy.
Subgoal
Intermediate goal defined by a higher-level agent that lower-level agents must achieve to progress toward the final goal.
Intra-option Policy
Policy that determines the actions to execute at each time step when a specific option is active within the hierarchical framework.
Feudal Networks (FuN)
Hierarchical architecture inspired by feudalism where a manager defines goal directions and workers execute actions to achieve these goals.
Controller
Lower-level agent that executes primitive actions to accomplish the subgoals specified by the meta-controller.
Hierarchical Deep Deterministic Policy Gradient (H-DDPG)
Extension of the DDPG algorithm incorporating a hierarchical actor-critic structure for learning in continuous action spaces.
Multi-level Actor-Critic
Architecture where each hierarchical level has its own actor-critic pair optimized for different temporal horizons.
Hierarchical Q-Learning
Q-learning variant where Q-values are computed at different hierarchical levels to evaluate options and primitive actions.
Subtask Decomposition
Process of automatically dividing a complex task into simpler, manageable subtasks for hierarchical learning.
End-to-end Hierarchical Learning
Approach where the entire policy hierarchy is trained simultaneously without manual pre-decomposition of tasks.