Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Bayesian Causal Inference
Application of the principles of Bayes' theorem to estimate causal effects from observational data, incorporating prior knowledge and quantifying the uncertainty of cause-and-effect relationships.
Residual Correlation Diagram (RCD)
A graphical tool used in causal inference to visualize conditional correlations between variables after accounting for their parents in a causal graph, helping to detect false edges.
Causal Score Equation
A mathematical formula that quantifies the contribution of a potential edge to the posterior probability of a causal graph, based on the structure and parameters of the Bayesian model.
Causal Structural Prior
A prior probability distribution over the space of possible causal graphs, incorporating expert knowledge or constraints to guide the search for the most plausible causal structure.
Bayesian Meta-Analysis of Causal Effects
A statistical approach that combines causal effect estimates from multiple studies using a Bayesian framework, allowing for accounting for heterogeneity between studies and obtaining an overall estimate of the causal effect.
Causal Latent Variable Models
An extension of causal Bayesian networks that includes unobserved (latent) variables that influence measured variables, used to model unmeasured confounding factors or theoretical constructs.
Posterior over Causal Graphs
A probability distribution over the set of possible causal graph structures after observing the data, representing the uncertainty about the true underlying causal structure.
Causal Decoupling
The principle that an intervention on a variable X (do(X)) statistically decouples X from its normal causes while preserving its relationship with its effects, fundamental for identifying causal effects.
Markovian Causal Equivalence
Concept according to which different causal graphs (partially oriented) can encode the same set of conditional dependencies and independencies, making their distinction impossible based solely on observational data.
Bayesian Confounding Factor
Variable that influences both the cause and effect, whose effect is quantified and adjusted in the Bayesian framework by integrating a probability distribution over its influence rather than a simple correction.
Probabilistic Intervention (Stochastic do-operator)
Generalization of the do-operator where the intervention on a variable does not fix it to a deterministic value but modifies its probability distribution, allowing modeling of policies or treatments with variable effects.
Bayesian G-Formula Method
Technique for estimating standardized causal effects that uses Bayesian adjustment to model the distribution of outcomes conditionally on covariates and exposure, allowing management of complex models and quantification of uncertainty.
Bayesian Causal Propensity
Extension of the propensity score where the probability of receiving treatment is modeled in a Bayesian framework, integrating uncertainty about the propensity model itself to improve causal adjustment.
Bayesian Causal Retro-analysis
Process of inferring probable causes of an observed effect by inverting causal reasoning using Bayes' theorem, calculating P(Cause|Effect) instead of simply P(Effect|Cause).
Score-based Causal Discovery
Family of algorithms that search for the causal graph maximizing a quality score (such as BIC or marginal probability) in a Bayesian framework, by exploring the space of possible structures.
Internal Causal Validity
In the Bayesian context, evaluation of the credibility of a causal relationship within a specific model, based on the posterior coherence of parameters and structure with observed data.