AI 용어집
인공지능 완전 사전
Options and Macro-actions
Extended temporary actions that encapsulate sequences of primitive actions, enabling temporal abstraction in hierarchical decision-making processes.
Hierarchical Reinforcement Learning (HRL)
Learning paradigm where policies are hierarchically structured with high-level options selecting sub-policies at lower levels.
MAXQ Decomposition
Hierarchical decomposition framework that represents the task as a directed acyclic graph of reusable subtasks with independent policies.
Teamwork and Hierarchy
Organizational structure where agents are grouped into teams with defined hierarchical roles to optimize collaboration and distributed decision-making.
HI-MAT (Hierarchical Multi-Agent Teaching)
Teaching method where expert agents transmit strategic knowledge to novices through a hierarchical structure of demonstrations.
Multi-level HRL
Extension of standard HRL with more than two hierarchical levels, allowing temporal abstractions at variable granularity.
Hierarchical Deep Q-Networks (hDQN)
Architecture combining deep Q-networks with hierarchical options, using high-level controllers and low-level sub-policies.
Goal-Conditioned Hierarchies
Hierarchical structures where policies at each level are conditioned by specific goals defined by the higher level.
Subtask Policies
Specialized policies optimized for accomplishing specific subtasks within the framework of a hierarchical decomposition of the overall problem.
Multi-Agent Hierarchical Reinforcement Learning (MAHRL)
Field integrating multi-agent challenges and hierarchical learning to solve complex distributed coordination problems.
Hierarchical Attention Networks
Neural network architecture applying attention mechanisms at multiple hierarchical levels to weight the importance of information.
Nested Hierarchies
Structures where hierarchies are nested within each other, allowing fine granularity in multi-level decision abstraction.
Temporal Abstraction in HRL
Fundamental principle of HRL enabling the grouping of action sequences into coherent temporal units to simplify planning.
Hierarchical Cooperative Networks
Networks of agents organized hierarchically where cooperation is structured according to defined levels of responsibility and authority.
Meta-Learning Hierarchies
Approach where agents learn to learn optimal hierarchical structures to quickly adapt to new tasks.
Hierarchical Graph Neural Networks
Extension of GNNs applied to multi-agent hierarchical structures, modeling dependencies across multi-level graphs.