|
Abstract:
|
The stochastic approximation Monte Carlo (SAMC) algorithm has been recently proposed in the literature as a general Monte Carlo and optimization algorithm. A remarkable feature of the algorithm is that it avoids the local-trap problem suffered by conventional MCMC algorithms; the algorithm can self-adjust the acceptance rate of the local Metropolis-Hastings moves such that each subregion of the sample space can be sampled with a desired frequency. In this talk, we will provide an overview of the algorithm from both the theoretical and practical aspects. Applications of the algorithm to a variety of computational problems, e.g., model selection, marginal density estimation, machine learning, phylogeny estimation, and global optimization, will be discussed.
|
