AI-ordlista
Den kompletta ordlistan över AI
Hill-Climbing Algorithm
Iterative optimization method progressively modifying the graph structure through local operations (edge addition, deletion, reversal) to improve the model's score.
Simulated Annealing
Stochastic optimization algorithm occasionally allowing solution degradations to escape local optima, inspired by the metallurgical cooling process.
D-Separation
Graphical criterion determining if two sets of variables are conditionally independent given a third set in a Bayesian network, based on path structure.
Bayes Factor
Ratio of marginal likelihoods comparing two competing graphical models, quantifying the evidence from the data in favor of each hypothetical structure.
Latent Structure
Graph configuration including unobserved (hidden) variables requiring specialized techniques like the EM algorithm for parameter and structure estimation.
Structural Cross-Validation
Robust evaluation method partitioning data to estimate the predictive capability of different graph structures, preventing structural overfitting.
Markov Chain Monte Carlo (MCMC)
Sampling technique exploring the space of structures according to their posterior probability, allowing approximation of the distribution over possible graphs.
Exact Structure Inference
Analytical calculation of the posterior distribution over graphical structures, mathematically exact but exponentially complex in practice for more than a few variables.
Moral Graph
Transformation of a directed graph into an undirected graph by connecting the parents of each node and removing the orientation, a prerequisite for certain inference methods.
Structural Complexity
A measure quantifying the informational richness of a graph by its number of edges and parameters, balanced against its predictive power in selection criteria.