YZ Sözlüğü
Yapay Zekanın tam sözlüğü
HRL (Hierarchical Reinforcement Learning)
Reinforcement learning paradigm that structures policies into hierarchical levels to solve complex tasks through temporal and spatial decomposition.
Semi-Markov Decision Process
Extension of the Markov decision process where transitions can take variable durations, naturally modeling long-term hierarchical actions.
Subtask Discovery
Automatic process of identifying and creating relevant subtasks to build an effective hierarchy without explicit human supervision.
Pareto Optimality
Concept where no solution can improve one objective without degrading another, forming the frontier of optimal solutions in multi-objective space.
Scalarization Functions
Functions that transform an objective vector into a single scalar value, enabling the application of single-objective algorithms to multi-objective problems.
Policy Gradient Methods for MO-HRL
Gradient-based policy optimization algorithms adapted to multi-objective hierarchical contexts, managing trade-offs between levels and objectives.
Value Function Decomposition
Technique that decomposes the global value function into contributions from each subtask and objective, facilitating distributed learning in hierarchies.
Intrinsically Motivated HRL
Approach where intrinsic motivations guide the discovery and selection of subtasks, improving exploration and efficiency in hierarchical learning.
Multi-Criteria Decision Making
Process of selecting actions or policies based on simultaneous evaluation of multiple quantitative and qualitative criteria within a hierarchical framework.
Objective Space Partitioning
Division of the objective space into regions managed by different hierarchical levels or specialized sub-policies for specific objective combinations.
Hierarchical Multi-Objective Policy Optimization
Simultaneous optimization of policies at multiple hierarchical levels aiming to maximize a set of often conflicting objectives with different time horizons.