CAJ | 학술논문

针对贪心最大割图半监督学习算法（简称GGMC）计算复杂度较高的问题，提出一种改进的贪心最大割图半监督学习算法（简称GGMC-Estop）。依据对GGMC算法优化过程中目标函数变化趋势的实验分析，采取两种在迭代初期停止GGMC算法运行策略，继而通过一次标准的标签传播步骤预测图上所有样本的标记来实施对GGMC的改进。典型数据集的仿真实验结果表明，在取得相近分类性能的同时，改进算法在计算速度上有很大的提高。
침대탐심최대할도반감독학습산법（간칭GGMC）계산복잡도교고적문제，제출일충개진적탐심최대할도반감독학습산법（간칭GGMC-Estop）。의거대GGMC산법우화과정중목표함수변화추세적실험분석，채취량충재질대초기정지GGMC산법운행책략，계이통과일차표준적표첨전파보취예측도상소유양본적표기래실시대GGMC적개진。전형수거집적방진실험결과표명，재취득상근분류성능적동시，개진산법재계산속도상유흔대적제고。
Aiming at the problem of high computational complexity of Greed Max-Cut Graph semi-supervised learning algorithm(GGMC), an improved Greed Max-Cut Graph semi-supervised learning algorithm based on Early stopping strategy, called GGMC-Estop, is proposed. According to the experimental analysis on that object function value in optimi-zation procedure of GGMC, the algorithm is improved in which two early stopping strategies are applied to stop GGMC training and prediction. Standard propagation is used to predict the label of data over the whole graph in one step. Experi-mental results on typical data sets show that the computational amount using the improved algorithm is far less than that of using GGMC algorithm, while the performance of classification for these two algorithms is almost approximate.