KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Metropolis-Hastings Method
MCMC sampling algorithm that generates a Markov chain whose stationary distribution matches a specified target distribution, using an acceptance-rejection criterion.
Markov Chain Monte Carlo (MCMC)
Class of algorithms for sampling from probability distributions by constructing a Markov chain that has the desired distribution as its equilibrium distribution.
Likelihood Weighting
Weighted sampling technique where weights are proportional to the likelihood of samples given the observations, used for inference in Bayesian networks.
Hamiltonian Monte Carlo Sampling
Advanced MCMC variant using Hamiltonian mechanics and gradients to propose efficient moves in parameter space, reducing sample autocorrelation.
Particle Filtering
Sequential Monte Carlo technique using a set of weighted particles to approximate probability distributions in dynamic state models and Bayesian filters.
Blocked Gibbs Sampling
Gibbs sampling variant where blocks of variables are sampled simultaneously rather than one variable at a time, improving mixing in certain configurations.
Robbins-Monro Algorithm
Stochastic search method for finding roots of a function when observations are noisy, fundamental in optimization and adaptive Monte Carlo methods.
Adaptive MCMC Sampling
Class of MCMC algorithms that adjust their parameters during execution to optimize acceptance rates and mixing properties of the Markov chain.
Coupling from the past
Monte Carlo technique to generate exact samples from the stationary distribution of a Markov chain by running backwards in time from an infinite past until coalescence.