Słownik AI
Kompletny słownik sztucznej inteligencji
Causal Reinforcement Learning
Learning paradigm that integrates principles of causal inference into RL algorithms to improve the generalization and robustness of learned policies in the face of environmental changes.
Counterfactual in RL
Reasoning about what would have happened if the agent had taken a different action in a given state, essential for unbiased value estimation in causally complex environments.
Directed Acyclic Causal Graph
Graphical structure representing causal dependency relationships between variables, where directed edges indicate the direct influence of one variable on another without cycles.
Structural Response Function
Mathematical function describing how a variable depends on its direct causes in a causal model, used to predict the effects of interventions in reinforcement learning.
Confounding Bias in RL
Systematic distortion of value estimation due to unobserved variables influencing both actions and rewards, which the causal approach seeks to correct.
Counterfactual Reward Distribution
Probabilistic distribution of rewards that would have been obtained under different actions, enabling more accurate estimation of policy values in causal environments.
Causal Meta-Learning
Approach learning to rapidly discover underlying causal structures in new environments to facilitate fast adaptation of reinforcement learning policies.
Causal Exploration
Exploration strategy that actively identifies causal relationships to maximize information acquired about the environment structure rather than simply maximizing immediate rewards.
Do-Calculus Equation
Set of formal rules allowing to transform expressions containing interventions (do()) into observable probabilities, essential for computing values in causal RL.
Causal Generalization
Ability of a learned policy to perform effectively in new environments sharing the same underlying causal structure, main objective of causal reinforcement learning.
Causal Latent Variables
Unobservable variables that exert causal influence on observable environment states, whose identification is crucial for policy robustness in causal RL.
Cross-Environment Transfer
Process of transferring knowledge learned in a source environment to target environments sharing common causal structures, facilitated by causal modeling.
Causal Robustness
Property of a reinforcement learning policy to maintain its performance in the face of variations in transition probability distributions, thanks to understanding causal relationships.
Relational Causal Reinforcement Learning
Extension of causal RL to environments with entities and relations, where the causal structure includes relational dependencies between environment objects.
Pearl's Principles
Theoretical foundations of causal inference including the causal hierarchy, structural models and do-calculus, applied to solve generalization problems in RL.
Causal Inference in RL
Process of identifying cause-effect relationships from agent-environment interaction data, enabling to distinguish correlation from causality in learning.