AI-ordlista
Den kompletta ordlistan över AI
Pearl Causal Hierarchy
Theoretical framework organizing causal inference into three levels: association (observation), intervention (action), and counterfactual (hypothetical reasoning), each requiring stronger assumptions.
Structural Causal Model (SCM)
Mathematical formalization combining structural equations, exogenous variables, and probabilistic assumptions to describe the causal mechanisms generating observed data.
Causal Bayesian Network
Bayesian network interpreted causally where edges represent direct causal relationships, allowing calculation of intervention effects and answering counterfactual questions.
Confounding Variable
Variable simultaneously influencing the cause and effect under study, creating a spurious association that must be controlled to isolate the true causal effect.
Counterfactual
Hypothetical question about what would have happened if a different decision had been made, requiring a causal model to estimate unobserved alternative outcomes.
Causal Effect
Change in a variable's distribution resulting from an intervention on another variable, distinct from the simple correlational association observed in the data.
Markov Blanket
Minimal set of variables making a variable conditionally independent of all other variables in the graph, composed of its parents, children, and co-parents.
Faithfulness Assumption
Assumption stating that all conditional independencies in the data derive from the causal graph structure, without accidental parameter cancellations.
Adjustment Set
Set of variables that can block all backdoor paths between treatment and outcome, satisfying the conditions for unbiased causal estimation.