Abstract
One of the most important roles of Kansei engineering is to understand and operate the human tacit knowledge. The neural network approach is very useful for the learning of tacit knowledge. The neural network approach makes the tacit knowledge operational, and we can use tacit knowledge effectively. Since the learned knowledge is embedded in the neural networks, the l earned knowledge is opaque and difficult to understand. And the leaned knowledge itself is difficult to evaluate. Hence the generation of if-then rules from neural networks is proposed, the generated rules are usually related to many cells, and the rules are not easy to understand. Especially, there are many hidden cells in BP algorithm, and the meanings of the hidden cells are difficult to identify. If we want to understand the learned knowledge from neural networks, we need some way of interpretation of the learned k nowledge to the formal knowledge. In this report, we apply revised learning algorithms from BP algorithm. Those algorithms are useful to clarify the relationships among the cells and the behaviors of hidden cells. And then we present a methodology of interpretation of the knowledge learned by the neural networks to the formal knowledge that can be easy to understand. We discuss the steps of knowledge interpretation from neural network, and show the Kansei engineering example of the knowledge interpretation, and discuss the role and the possibility of the knowledge interpretation.
