| 摘要: |
| 本文提出了一个综合多种神经网络理论的学习算法ＦＴＡＲＴ（ｆｉｅｌｄｔｈｅｏｒｙ—ｂａｓｅｄａｄａｐｔｉｖｅｒｅｓｏｎａｎｃｅｔｈｅｏｒｙ），它将ＡＲＴ（ａｄａｐｔｉｖｅｒｅｓｏｎａｎｃｅｔｈｅｏｒｙ）算法、ＦｉｅｌｄＴｈｅｏｒｙ和ＡＲＴＭＡＰ等算法的优点有机结合，并以样本在实例空间出现的概率为启发信息修改分类．ＦＴＡＲＴ由于采用了不同于其它算法的冲突解决和动态扩大分类区域的方法，因此取得了较好的效果．本文还对实现ＦＴＡＲＴ算法的结果进行了验证． |
| 关键词: 神经网络 自适应谐振理论 域理论 |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家自然科学基金资助. |
|
| A NEW ADAPTIVE RESONANCE THEORY ALGORITHM |
|
Chen Zhaoqian,Zhou Rong,Liu Hong,Chen Shifu
|
| Abstract: |
| This paper proposes a neural network learning algorithm FTART(field theory-based adaptive resonance theory)that combines many neural network theories.It is based on the main idea of ART(adaptive resonance theory) and makes use of many merits of Field Theory and ARTMAP,etc.,and corrects the classification regarding the supposed possibility of examples in the instance space.FTART gains some goodness by employing a special conflict resolution process and dynamically expanding the classification area. Some benchmark test results are also given. |
| Key words: Neural network adaptive resonance theory field theory. |
