论文部分内容阅读
用一组训练样本对神经网络进行训练后,网络对训练阶段未曾见过的样本也能正确分类。但传统的神经网络模式分类方法泛化能力不十分理想,而且不稳定。对同一个分类任务,训练样本改变,分类器泛化能力的大小也会改变。该文提出一种基于最优分类面的神经网络模式分类方法。通过寻找并训练最优分类面,提高网络的泛化能力,增强泛化能力的稳定性。用异或问题和双螺旋线问题验证该新方法的有效性和泛化能力,取得了令人满意的结果。
After training a neural network with a set of training samples, the network can correctly classify the samples that have not been seen in the training phase. However, the generalization ability of the traditional neural network pattern classification method is not very ideal and unstable. For the same classification task, the training sample changes, the size of classifier generalization ability will also change. This paper proposes a neural network pattern classification method based on the optimal classification surface. By finding and training the optimal classification surface, the generalization ability of the network is improved and the stability of the generalization ability is enhanced. Validation of the validity and generalization ability of this new method by using XOR problems and double helix problems has yielded satisfactory results.