KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Structural Causal Model
Mathematical framework combining directed acyclic graphs and structural equations to represent and analyze causal relationships between variables in a system.
Directed Acyclic Graph (DAG)
Graphical representation where nodes are variables and directed edges indicate direct causal influences, without cycles to ensure causal consistency.
Pearl's Ladder of Causation
Three-level hierarchy of causal reasoning: association (correlation), intervention (manipulation), and counterfactual (imagination of alternative worlds).
Causal Identification
Mathematical process determining whether a causal effect can be uniquely estimated from observational data and structural assumptions.
Structural Equation Model (SEM)
Statistical framework using simultaneous equations to model causal relationships and errors between latent and observed variables.
Intervention Distribution
Probability distribution resulting from an external intervention on the system, different from the observational distribution as it modifies structural relationships.
Linear Causal Models
Class of structural models where each variable is a linear combination of its direct causes plus an independent error term.
Causal Falsifiability
Ability of a causal model to generate testable and refutable predictions from structural assumptions, distinguishing science from speculation.
Admissibilité (Admissibility)
Condition selon laquelle un ensemble de variables est suffisant pour ajuster l'estimation d'un effet causal sans introduire de biais de sélection.
Graph Causal Bayesian Network
Réseau bayésien augmenté d'une interprétation causale où les arêtes représentent des mécanismes causaux plutôt que de simples dépendances probabilistes.
Modèles de Réponse Potentielle
Framework alternatif modélisant les résultats potentiels pour chaque unité sous chaque traitement possible, fondement de l'inférence causale en statistiques.
Causal Discovery
Ensemble d'algorithmes visant à inférer la structure causale (graphe et équations) à partir de données observationnelles ou expérimentales.