基于独立成分分析的分解向前SVM降维算法

提出一种基于大样本学习的分解向前支持向量机算法和一种新的基于独立成分分析的降维学习模型，其算法的复杂度比传统块算法和标准SVM低。利用不完备ICA思想，达到数据压缩而降维的目的。实验发现，由于降低了输入维数，简化了数据结构，从而减少了SVM识别的计算复杂度，当把向量维数从110维降低到5维时，平均识别率超过传统神经网络达到93%，因而从计算时间和识别效率二者的综合情况来考虑，ICA降维模型是一种理想的实际应用模型。

Decomposition forward SVM dimension-reduction algorithm based on independent component analysis

A Decomposition Forward Support Vector Machine (DFSVM) algorithm for large scale samples learning and a new dimension reduction model based on Independent Component Analysis (ICA) were proposed. The calculational complexity is lower than that of the traditional chunking algorithm and the standard SVM. Use the idea of imcomplete ICA to compress data and reduce the dimension. Because of the reduced input dimension and simplified data structure, the calculational complexity of SVM has been reduced. Experiment indicates that if reducing the dimension from one hundred and ten dimension to five-dimension, the average recognition rate is superior to traditional neural network and comes to 93%. Considering the time cost and the recognition efficiency, ICA dimension reduction model is an ideal application model in practice.