Plausible Neural Network - A self-organized Network for Large-scale Multivariate Data Analysis

Abstract

There are several difficulties for developing a large-scale multivariate data analysis method. Some issues are statistical concern: for example redundancy/collinearly among variables, noisy and missing data value, mixture of categorical and quantitative data with different measurements scale, restriction of parameters due to the shortage of sample size. Another issue is computational concern due to the curse of dimensionality. Plausible Neural Network (PNN) is a new hybrid intelligent system, developed by the speaker to meet the challenge. PNN combine neural network, statistical inference and fuzzy set theory into a single framework. PNN has very fast learning algorithm with O(N) computation complexity. Currently it can handle more than 30,000 variables with a desktop machine. In this talk, I will introduce PNN, no background knowledge of neural network or fuzzy set theory is required. The closest resembling model of the PNN is the Bayesian network; however, PNN is more general. I will demonstrate the application of PNN to some high dimensional bioinformatics data analysis problems, such as gene expression analysis and proteomic pattern recognition.