欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
为了处理自训练半监督支持向量机算法中每次循环都需要求解二次规划因此效率低的问题,采用直接求解支持向量机的原始优化问题,由此得到一个不光滑的无约束优化问题。将正号函数展开为无穷多项式级数,由此得到了一族光滑函数,用多项式光滑函数对无约束优化问题进行逼近,并用共轭梯度算法求解模型。在人工数据和UCI数据集上的实验结果显示,给出的算法效率高,能保证标记样本很少时的分类精度并且不因标记样本的增多而明显提高分类精度。
英文摘要：
In order to deal with the problem of low efficiency brought by solving the quadratic programming in every cycle of self-training semi-supervised support vector machines, the method of training support vector machine in the primal was used. Therefore a nonsmooth optimization problem without constraint was deduced. Plus function was transformed to an equivalent infinite series. Thus a family of smoothing functions was derived. The polynomial smoothing functions was used to approach the optimization problem without constraint and the conjugate gradient algorithm was used to solve the model. Experimental results on artificial and real data support that the proposed algorithm with higher efficiency can guarantee the accuracy when the percentage of labeled sample is very low and the accuracy does not improve obviously as the number of labeled data increasing.
同期刊论文项目
同项目期刊论文
期刊信息
位置：