AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Preference-based Reinforcement Learning
Approach where the agent learns from comparisons between different trajectories, without requiring explicit numerical rewards.
Reward Model from Comparisons
IRL technique that constructs a reward function by analyzing user preferences expressed during pairwise comparisons of actions or trajectories.
Feedback-based Reinforcement Learning
Paradigm where the agent continuously adjusts its policy by integrating qualitative and quantitative corrections provided by the user.
Active Learning for IRL
Strategy where the agent actively selects the most informative questions or demonstrations to minimize uncertainty about the reward function.
Cooperative Inverse Reinforcement Learning
Method where the user and agent actively collaborate, with the user providing guided corrections and the agent proposing iterative improvements.
Bayesian Reward Function
Probabilistic approach that models uncertainty about the reward function and updates beliefs as new information is received.
Multi-Objective Inverse Reinforcement Learning
Extension of IRL where multiple conflicting reward functions must be discovered and weighted simultaneously.
Deep Inverse Reinforcement Learning
Use of deep neural networks to represent complex, non-linear reward functions from human demonstrations.
Online Inverse Reinforcement Learning
Variant where the agent learns and adjusts the reward function in real-time during interaction with the environment and user.
Reward Reinforcement Inverse Reinforcement Learning
Iterative process where the reward function is progressively refined through cycles of feedback collection and model improvement.
Transfer Learning Inverse Reinforcement Learning
Technique that leverages knowledge acquired in previous tasks to accelerate the learning of new reward functions.
Contextual Inverse Reinforcement Learning
Approach where the reward function depends on the context or state of the environment, allowing for conditional preferences.
Inverse Reinforcement Learning for Complex Systems
Application of IRL to environments with large state and action spaces, requiring advanced approximation techniques.
Continual Learning Inverse Reinforcement Learning
Framework where the agent continuously adapts to changes in user preferences without forgetting previously acquired knowledge.
Trajectory Similarity Metric
Function quantifying the resemblance between different agent trajectories, used to evaluate compliance with human preferences.