Improving Efficiency in Parameter Estimation Using the Hamiltonian Monte Carlo Algorithm

摘要: Faculty of Mathematics and Natural Sciences Department Informatics Master Science by Mohammed Alfaki The Hamiltonian Monte Carlo algorithm, or alternately called hybrid is Markov chain technique, which combines a Gibbs sampling update with the Metropolis acceptance-rejection rule. algorithm simulates distribution using dynamics which, involves gradient information to investigate space, thus has better convergence properties than Metropolis–Hastings algorithms. suffers from random walk in generating momentum, an additional error when simulated constant step–size. This thesis investigates three approaches improve performance algorithm. first approach enhances suppressing ordered over–relaxation. second simulation adaptive step–size reduce simulation. third proposal combine two versions into one