CAJ | 학술논문

　　当训练数据每次发生改变时,例如增加或者删除部分数据,标准支持向量机的批处理算法就需要重新进行训练,这将不适合在线环境的计算。为了克服这个问题,Cauwenberghs 和 Poggio 提出了增量和减量式标准支持向量机算法(C&P算法)。通过理论分析,证明C&P算法的可行性和有限收敛性。可行性证明确保了C&P算法的每步调整都是可靠的,有限收敛性证明确保了 C&P 算法通过有限步调整最终收敛到问题的最优解。在此基础上,进一步通过实验结果验证了所给出的理论分析的结果。
　　당훈련수거매차발생개변시,례여증가혹자산제부분수거,표준지지향량궤적비처리산법취수요중신진행훈련,저장불괄합재선배경적계산。위료극복저개문제,Cauwenberghs 화 Poggio 제출료증량화감량식표준지지향량궤산법(C&P산법)。통과이론분석,증명C&P산법적가행성화유한수렴성。가행성증명학보료C&P산법적매보조정도시가고적,유한수렴성증명학보료 C&P 산법통과유한보조정최종수렴도문제적최우해。재차기출상,진일보통과실험결과험증료소급출적이론분석적결과。
Batch implementations of standard support vector machine (SVM) are inefficient on an online setting because they must be retrained from scratch every time the training set is modified (i.e., adding or removing some data samples). To solve this problem, Cauwenberghs and Poggio propose an incremental and decremental support vector classification algorithm (C&P algorithm). This paper proves the feasibility and finite convergence of the C&P algorithm through theoretical analysis. The feasibility ensures that each adjustment step in the C&P algorithm is reliable, and the finite convergence ensures that the C&P algorithm can converge to the optimal solution within finite steps. Finally, the conclusions of the theoretical analysis are verified by the experimental results.