Glosarium AI
Kamus lengkap Kecerdasan Buatan
Hierarchical Imitation Learning
Learning paradigm where an agent learns complex behaviors by decomposing expert demonstrations into a hierarchy of interconnected subtasks.
Hierarchical Policy
Multi-level control structure where high-level policies select low-level sub-policies to sequentially accomplish intermediate goals.
Atomic Subtasks
Fundamental and indivisible action units in the task hierarchy, which cannot be further decomposed and serve as basic building blocks for learning.
Dependency Graph
Formal structure representing precedence relationships and constraints between different subtasks in a hierarchical decomposition.
Multi-scale Imitation
Learning approach where the agent simultaneously imitates behaviors at different levels of temporal granularity and spatial abstraction.
Demonstration Sequence
Ordered chain of hierarchically structured expert examples, capturing both macro-actions and micro-movements necessary for task completion.
Learning by Composition
Learning method where new skills are acquired by creatively combining previously learned sub-skills through imitation.
Trajectory Abstraction
Process of generalizing raw demonstrations into abstract behavioral patterns, capturing underlying intention rather than execution details.
High-Level Policy
Decision-making strategy operating at a high level of abstraction, responsible for selecting and sequencing sub-goals to achieve.
Low-Level Policy
Detailed motor controller implementing concrete action primitives necessary for executing sub-tasks specified by the high-level policy.
HAM Formalism
Mathematical framework (Hierarchical Abstract Machines) defining hierarchical state machines to structure imitation learning policies.
Hierarchical Planning
Process of generating multi-level action plans where strategic decisions recursively decompose into more detailed tactical plans.
Transferable Imitation Learning
Ability of a system to transfer knowledge acquired through hierarchical imitation between tasks sharing similar decomposition structures.
Hierarchical Demonstrations Model
Probabilistic representation capturing the joint distribution of actions and sub-goals across different levels of abstraction in expert demonstrations.
Imitation Learning with Primitives
Approach where complex behaviors are learned by automatically identifying and combining fundamental movement primitives extracted from demonstrations.
Hierarchical Neural Networks
Deep learning architecture organized in specialized layers, each level processing representations at different scales of abstraction for imitation.
Multi-level Temporal Alignment
Technique for synchronizing demonstrations at different temporal scales to ensure coherence between high and low level policies.