Trust Learning and Estimation

Trust Learning and Estimation

Gehao Lu (University of Huddersfield, UK and Yunnan University, China) and Joan Lu (University of Huddersfield, UK)
Copyright: © 2017 |Pages: 14
DOI: 10.4018/978-1-5225-1884-6.ch018

Abstract

Predict uncertainty is critic in decision making process, especially for the complex systems. This chapter aims to discuss the theory involved in Self-Organizing Map (SOM) and its learning process, SOM based Trust Learning Algorithm (STL), SOM based Trust Estimation Algorithm (STL) as well as features of generated trust patterns. Several patterns are discussed within context. Both algorithms and how they are processed have been described in detail. It is found that SOM based Trust Estimation algorithm is the core algorithm that help agent make trustworthy or untrustworthy decisions.
Chapter Preview
Top

2. Theory Of Self Organizing Map

2.1 Self Organizing Map

Self-organizing map (SOM) is one type of neural networks based on competitive learning (Kohonen, 1998). The features of SOM include: approximation of input spaces, topologically ordering, density matching and feature selection (Haykin, 1999). The belief of trust is frequently mutating and updating, the chosen algorithm should be highly adaptive. This is what SOM is capable of. The belief of trust is a continuous accumulation of information. The output of SOM is compact and completes enough to reflect every piece of changes. Also, the output of SOM can reveal obvious characteristics of the trust belief. It is easy for a computer program to discover the belief of trust which comes from multiple sources.

The SOM can approximate the input space. The statistical characteristics can be extracted from the input data. The result map reveals an approximation of the input space through the synaptic weight of the output space. The SOM can discover the topological ordering of the input data. The location of the excited neurons constructs a spatial pattern that represents the geometric characteristics of the input space. Because the neighboring neurons are also excited, the zone that reflects the input space has better chances to be selected. This results in a good density matching. Thus, the features hided in the input space can be found through discovering the special distribution of the neuron groups.

The basic process of SOM is made up by three processes: competitive process, cooperative process and adaptation process. After initializing the synaptic weights of the network, in the competitive process, the data of the input pattern are provided to the discrimination function of the network, for each input pattern at each neuron, there will be a value of the discrimination function. The highest value gained neuron will be the winner of the competition. In the cooperative process, the neighbor function decides which neighbor neurons around the winning neuron are excited as well. In the adaptation process, the synaptic weights are adjusted so that the similar pattern leading to the winning neuron will be enhanced.

Complete Chapter List

Search this Book:
Reset