Study on the Structure and Behavioral Choice of IDs Model

摘要

The complicated decision making problem is one of the important components for the study on the system of artificial intelligence area. This thesis, based on the Bayesian technology and decision-making theory, is going to optimize the traditional IDs model and improve the ability of expression of the model. Firstly, the structure decomposition method is given to divide the IDs into two parts, which are probability network structure and utility function structure. Secondly, a new MDL evaluation standard is put forward to reduce the dependence on statistics of the traditional MDL evaluation standard and based on the new standard to propose to use the PS-EM in the model choice of probability network structure; and also by using the sum of individual utility function instead of the joint utility function to create the BP neural network to study the utility function structure of the IDs. The experimental result shows the method mentioned above is effective.