Fuzzy classification is the task of partitioning a feature space into fuzzy classes. A learn-by-example mechanism is desirable to automate the construction process of a fuzzy classifier. In this paper we introduce a method of employing adaptive networks to solve a fuzzy classification problem. System parameters, such as the membership functions defined for each feature and the parameterized t-norms used to combine conjunctive conditions are calibrated with backpropagation. We also employ Kalman filter algorithm to improve the overall performance. To explain this new approach, first we introduce the concept of adaptive networks and derive a supervised learning procedure based on a gradient descent algorithm to update the parameters in an adaptive network. Next, we discuss the use of Kalman filter algorithm to minimize the square error. From the experiment results, it is summarized that the adaptively adjusted classifier performs well on an Iris classification problem. The results are discussed from the veiwpoint of feature selection.
|Number of pages||4|
|State||Published - 1 Dec 1992|
|Event||Proceedings of the 1992 Japan - USA Symposium on Flexible Automation Part 2 (of 2) - San Francisco, CA, USA|
Duration: 13 Jul 1992 → 15 Jul 1992
|Conference||Proceedings of the 1992 Japan - USA Symposium on Flexible Automation Part 2 (of 2)|
|City||San Francisco, CA, USA|
|Period||13/07/92 → 15/07/92|