Gibbs sampling for Dirichlet process Gaussian mixture models
March 21st, 2016
An algorithm for learning Dirichlet Process Mixtures using Gibbs sampling will be presented. It will also be described how Dirichlet Process Mixtures can be used to construct optimal perturbation kernels in Approximate Bayesian Computation using Sequential Monte Carlo (ABC-SMC) approaches. The adapted ABC-SMC will be used to estimate the parameters of a model of a human signalling pathway, the Wnt pathway, using real experimental data. Finally, it will be shown, based on the results, that the adapted kernels increase the performance of ABC-SMC and that the added computational cost of adapting kernel functions is easily regained in terms of the higher acceptance rate.