Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Matérn Kernel
Family of covariance functions parameterized by a smoothness factor ν, offering flexible control over the differentiability of the Gaussian process.
Kernel Hyperparameters
Parameters of the covariance function (e.g., length-scale, signal variance) that control the behavior of the Gaussian process and are typically learned by maximizing the marginal likelihood.
Gaussian Process Posterior
Distribution of the Gaussian process conditioned on observed data, combining the prior and the likelihood to provide predictions with uncertainty.
Length-Scale
Kernel hyperparameter that determines the distance over which input points are correlated, controlling the smoothness of the modeled function.
Gaussian Process Prediction
Bayesian inference that calculates the mean and variance of the predictive distribution at new points, based on the Gaussian process posterior.
Kernel Regression
Non-parametric regression method using kernel functions, of which Gaussian process regression is a specific Bayesian case.
Gaussian Process Classification
Application of Gaussian processes to classification tasks, using link functions (e.g., probit) to model class probabilities.
Inducing Points
Latent variables in sparse Gaussian processes that summarize the information from the training data to speed up computations.
Periodic Kernel
Covariance function designed to model periodic patterns in data, essential for time series with cycles.
White Noise Kernel
Covariance function that models independent white noise, used to represent uncorrelated random uncertainty in observations.
Composite Kernel
Combination of multiple kernels (by addition or multiplication) to capture complex and multi-scale correlation structures in data.
Spatio-Temporal Gaussian Process
Extension of Gaussian processes to model data dependent on both space and time, using separable or non-separable spatio-temporal kernels.
Variational Inference for GP
Variational inference method applied to Gaussian processes to approximate the posterior while reducing computational complexity.