摘要
文中提出了一种具有抗噪音能力的增量式混合学习算法ＩＨＭＣＡＰ．该算法将基于概率论的符号学习与神经网络学习相结合，通过引入ＦＴＡＲＴ神经网络，不仅实现了两种不同思维层次的靠近，还成功地解决了符号学习与神经网络学习精度之间的均衡性问题．其独特的增量学习机制不仅使得它只需进行一遍增量学习即可完成对新增示例的学习，还使该算法具有较好的抗噪音能力，从而可以应用于实时在线学习任务．
In the paper here, an incremental hybrid learning algorithm IHMCAP is proposed, which has the ability of noise resistance. This algorithm successfully combines probability based symbolic learning with neural learning. By adopting the FTART neural network proposed before, the IHMCAP not only proportions the learning accuracy between symbolic and neural parts, but also lays the two different thought levels aboard. The unique incremental learning mechanism employed enables the IHMCAP performs only one round learning to cover new training patterns. Moreover, it also depresses the noise sensibility of the learning system, which makes IHMCAP fit for tasks that require real time online learning.
出处
《计算机研究与发展》
EI
CSCD
北大核心
1999年第6期675-680,共6页
Journal of Computer Research and Development
基金
国家自然科学基金
江苏省自然科学基金
关键词
神经网络
抗噪音
混合学习
机器学习
算法
hybrid model, incremental learning, neural network, noise disposal
