ART Network Robustness Enhancement Method Research

Artificial Neural Network(ANN) is an artificial imitation of the human brain signal processing system for processing the signal processing procedure. Its ability of records and processing information is achieved by changing the weight of connections between the artificial neurons. Artificial neural network is a highly non-linear signal processing system, and it has a good parallel processing capability and anti-noise performance. Artificial neural networks have mainly used in learning, classification, prediction and other fields, and have been developed for a variety of artificial neural networks to adapt the different application environments. Wherein, the adaptive resonance theory model(ART) gets the majority of researchers favor because of its good performance on Stability- Plasticity Problems.The adaptive resonance theory model 2(ART2) is the most representative network of the ART network family, and current ART network’s applications and research is also focus on the ART2 network. Therefore, this article also focus on the ART2 network, when it try to strengthen the ART network’s robustness. ART2 network inherits all the benefits of ART network, but there are still many problems which affect robustness of ART2 network, like input amplitude information is lost, the model drifting and noise pollution. This mainly study the robustness enhancement of ART networks, and the research object is the three aspects of the ART network, such as the model drift, the class division and the noise. Delay correction algorithm is put forward to slow down the model drift; introduction the clustering theory to solve the classification problem; use Ebbinghaus’ s memory forgetting curve is introduced to solve the noise problem.Firstly, this paper analysis the ART network for large data,which start with big data analysis of the Internet. The analysis show that the ART Network has a good practical significance argument. Then,further study the structure and principles of traditional ART network algorithm, analysis of the current status of research in the theory and its advantages and disadvantages. Furthermore, this paper focus on three problems: the influences of mode drift to the ART2 network, the influences of the input order to the ART2 network, the influences of noise to the result of ART2. And proposed the "prevention, management with emphasis on prevention," improvement strategy to solve the ART2 network’s model drift problem and the inputs order problem; cited Ebbinghaus memory- forgetting curve to improve the anti-noise ability of ART2 network. Besides, the paper also write codes by MATLAB to verify the performance of the new ART2 network. At last, summarized the works and the innovations of this paper, and pointed out the direction for the future research.