构建了一种通过DBN实现的1/4超球面支持向量机(Quarter-Sphere support vector machines,QSSVM)测试模型,设计了一种可以实现在线测试功能的异常检测算法。当窗口扩大后,QSSVM发生了准确度不断提高的变化趋势,能够提高半径的测试精度。...构建了一种通过DBN实现的1/4超球面支持向量机(Quarter-Sphere support vector machines,QSSVM)测试模型,设计了一种可以实现在线测试功能的异常检测算法。当窗口扩大后,QSSVM发生了准确度不断提高的变化趋势,能够提高半径的测试精度。当窗口增大后算法持续时间增加,QSSVM相对于OCSVM(One-Class support vector machines,OCSVM)可以降低近一半的计算时间。随着窗口扩大至临界值后,将会引起准确度的减小,设置QSSVM算法滑动窗口为100。当样本包含更高的维度异常比例时,所有算法都出现了检测率增大的变化现象。当样本维度升高后,QSSVM依然具备优异检测性能,而K-means发生了检测性能的下降。通过测试发现,采用新算法能够满足大规模高维传感器的数据处理需求,从而减小时间复杂度并能够更加准确测定异常数据。展开更多
The hypersphere support vector machine is a new algorithm in pattern recognition. By studying three kinds of hypersphere support vector machines, it is found that their solutions are identical and the margin between t...The hypersphere support vector machine is a new algorithm in pattern recognition. By studying three kinds of hypersphere support vector machines, it is found that their solutions are identical and the margin between two classes of samples is zero or is not unique. In this letter, a new kind of hypersphere support vector machine is proposed. By introducing a parameter n(n>1), a unique solution of the margin can be obtained. Theoretical analysis and experimental results show that the proposed algorithm can achieve better generaliza-tion performance.展开更多
文摘构建了一种通过DBN实现的1/4超球面支持向量机(Quarter-Sphere support vector machines,QSSVM)测试模型,设计了一种可以实现在线测试功能的异常检测算法。当窗口扩大后,QSSVM发生了准确度不断提高的变化趋势,能够提高半径的测试精度。当窗口增大后算法持续时间增加,QSSVM相对于OCSVM(One-Class support vector machines,OCSVM)可以降低近一半的计算时间。随着窗口扩大至临界值后,将会引起准确度的减小,设置QSSVM算法滑动窗口为100。当样本包含更高的维度异常比例时,所有算法都出现了检测率增大的变化现象。当样本维度升高后,QSSVM依然具备优异检测性能,而K-means发生了检测性能的下降。通过测试发现,采用新算法能够满足大规模高维传感器的数据处理需求,从而减小时间复杂度并能够更加准确测定异常数据。
基金Supported by the National Natural Science Foundation of China (No.60277101, No.60301003, No.60431020), Beijing Foundation (No.3052005), and Beijing Munici-pal Commission of Education Project (KM200410005030).
文摘The hypersphere support vector machine is a new algorithm in pattern recognition. By studying three kinds of hypersphere support vector machines, it is found that their solutions are identical and the margin between two classes of samples is zero or is not unique. In this letter, a new kind of hypersphere support vector machine is proposed. By introducing a parameter n(n>1), a unique solution of the margin can be obtained. Theoretical analysis and experimental results show that the proposed algorithm can achieve better generaliza-tion performance.
