基于认知几何的支持向量机分类

Support Vector Machine for Classification Based on Cognitive Geometry

支持向量机(SVM)较好地解决了小样本分类问题,但仍然受稀疏数据和噪音的影响.鉴于人类具有很好的处理稀疏数据和噪音问题的能力,文中提出了模型化这些认知能力的几何化方法,特别是采用相对变换方法建立了认知相对性规律的几何化模型,并用之改进了SVM.仿真实验结果表明,改进的SVM明显提高了抵抗稀疏数据和噪音的能力.

Although a support vector machine （SVM） has excellent classification ability for small data sets, it is still inefficient for noisy or sparse data sets. As humans can effectively deal with noisy and sparse data, a geometric approach of modeling human＇s cognitive abilities is proposed in this paper. Moreover, a geometric model of the relative cognitive law is established via relative transformation and is then used to improve SVM. It is indicated from the simulation that the classification capability of the improved SVM for noisy and sparse data sets significantly increases.