Interval propagation reasoning method of Ising graphical model

Abstract

The present invention discloses an interval propagation reasoning method based on Ising graphical model, wherein the method mainly comprises the following steps: firstly establishing an Ising mean field lopping computation tree based on Ising graphical model; and using a mean field interval propagation algorithm based on the Ising mean field lopping computation tree for computing the expected boundary of root node variate. The reasoning method of the invention has lower computing complexity and has an effective reasoning precision through the expected boundary of variate.

Description

technical field [0001] The invention relates to an approximate probability reasoning method on a graph model, especially an approximate variation reasoning method on an Ising graph model. Background technique [0002] 1. Ising graph model [0003] The Ising graphical model (Ising graphical model) originated in statistical physics and is a Markov random field model based on binary random vectors. It provides an important modeling method for image analysis and natural language processing and other fields. It is a probabilistic model built on a graph structure G = (V, E), where the node set V corresponds to a Bernoulli random vector x = {x 1 ,...,x n} ∈ {0, 1} n , the edge set E corresponds to the conditional independence relationship between variables. The exponential family probability density distribution p(x, θ) of the Ising graph model is [0004] p ( x ; ...

AI Technical Summary

Problems solved by technology

The local training method uses the initial few belief message iterations to calculate the model partition function, and the model parameters can be trained separately under the local information, which reduces the computational complexity and facilitates the addition of new information, but the local training method is difficult to measure the approximate belief propagation. calculation accuracy

The BP-SAW method performs finite belief propagation on the graph model SAW (self-avoiding walk) calculation tree, calculates the approximate distribution of marginal probability, and gives the error bound of the approximate distribution based on the concept of message error, but this method needs to be used in the entire model For message propagation, the computational complexity is high

[0029] These existing approximate variational inference methods are mainly based on belief propagation for approximate calculation research, and rarely consider other variational inference methods, such as the mean field inference method; at the same time, calculation accuracy is an important indicator of approximate variational inference research, and these methods Or it is difficult to analyze the calculation accuracy, such as the local training method, or the calculation complexity is high, such as the BP-SAW algorithm

Images