AI Glossary
The complete dictionary of Artificial Intelligence
Deviance Information Criterion (DIC)
Bayesian model selection metric that evaluates model fit while penalizing complexity, analogous to AIC in frequentist statistics.
Watanabe-Akaike Information Criterion (WAIC)
Fully Bayesian model selection method estimating the expected divergence between the model and the true generating process.
Bayes Score
Logarithm of the marginal probability of the data, used to rank models in Bayesian model selection.
Latent Space Model
Class of Bayesian models where unobserved latent variables are inferred to explain the structure of observed data.
Reversible-Jump MCMC Method
Advanced sampling algorithm that allows exploring models of different dimensions for Bayesian model selection.
Zellner-Siow Prior
Default prior distribution for Bayesian linear regression, favoring parsimonious models in variable selection.
Bayesian Model Averaging (BMA)
Technique combining predictions from multiple models weighted by their posterior probabilities to improve inferential robustness.
Posterior Model Probability (PMP)
Probability that a given model is the true generative model of the data, after Bayesian updating with observations.
Bayesian Cross-Validation (BCV)
Model validation method that evaluates its predictive ability by partitioning the data and comparing predictions to observations.
Bayes Factor Test
Decision-making procedure based on the Bayes factor to accept or reject a null hypothesis in a Bayesian context.
Bayesian Variable Selection
Application of Bayesian model selection to identify relevant predictors in regression models.
g-prior
Family of prior distributions for Bayesian hypothesis testing, controlling the distance between the null and alternative hypotheses.
Kass-Raftery Scale
Standardized interpretation of Bayes factors into categories (weak, moderate, strong, very strong) to assess the significance of evidence.
Bayesian Non-Parametric Model
Bayesian approach where the number of parameters can grow with the data, using processes like Dirichlet or Gaussian.
Variational Approximate Inference
Method for approximating the posterior distribution for model selection when exact calculations are intractable.