Copyright Statement: This is an open access article licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, even commercially as long as the original work is properly cited.
Digital Object Identifier (DOI) : 10.14569/IJARAI.2013.020706
Article Published in International Journal of Advanced Research in Artificial Intelligence(IJARAI), Volume 2 Issue 7, 2013.
Abstract: In this paper, we propose a new type of informationtheoretic method. We suppose that a neuron should be evaluated from different points of view to precisely discern its properties. In this paper, we restrict ourselves to two types of evaluation methods for neurons, namely, self and outer-evaluation. A neuron fires only as a result of evaluating itself, while the neuron can fire as a result of evaluation by all surrounding neurons. Selfand outer-evaluation should be equivalent to each other. When contradiction between two types of evaluation exists, the contradiction should be as small as possible. Contradiction between self- and outer-evaluations is realized in terms of the Kullback- Leibler divergence between two types of neurons. Contradiction between self- and outer-evaluation can be resolved by decreasing the contradiction ratio between the two types of evaluation in terms of KL divergence. This method is expected to extract the main features in input patterns, if those are shared by two types of evaluation. We applied the method to two data sets, namely, the logistic and dollar-yen exchange rate data. In both problems, experimental results showed that visualization performance could be improved, leading to clearer class structure for both problems. In addition, when visualization was improved, generalization performance did not necessarily degrade, showing the possibility of networks with better visualization and prediction performance.
Ryotaro Kamimura, “Contradiction Resolution between Self and Outer Evaluation for Supervised Multi-Layered Neural Networks” International Journal of Advanced Research in Artificial Intelligence(IJARAI), 2(7), 2013. http://dx.doi.org/10.14569/IJARAI.2013.020706