General-purpose use of Bayesian learning for hierarchical probabilistic models

本计画的目的是建立一个通用的使用贝叶斯学习的阶层式机率模型，例如一个类神经网路和一个隐藏的马可夫模型。我们使用一个交换蒙特卡罗方法进行贝叶斯学习效率，并显着提高计算成本通过并行化的exhcnage蒙特卡罗方法。此外，我们构造了一个交换蒙特卡罗方法的优化设计，并将所提出的方法应用于径向基函数网络的谱反卷积。

The purpose of this project is to establish a general-purpose use of Bayesian learning for hierarchical probabilistic model such as a neural network and a hidden Markov model. We use an exchange Monte Carlo method for performing Bayesian learning efficiently, and significantly improve the computational cost by parallelizing the exhcnage Monte Carlo method. Moreover, we construct a optimal design of the exchange Monte Carlo method and apply the proposed method to the spectral deconvolution for the radial basis function networks.