论文部分内容阅读
支持向量数据描述(SVDD)是一种无监督学习算法,在图像识别和信息安全等领域有重要应用.坐标下降方法是求解大规模分类问题的有效方法,具有简洁的操作流程和快速的收敛速率.文中针对大规模SVDD提出一种高效的对偶坐标下降算法,算法每步迭代的子问题都可获得解析解,并可使用加速策略和简便运算减少计算量.同时给出3种子问题的选择方法,并分析对比各自优劣.实验对仿真和真实大规模数据库进行算法验证.与LibSVDD相比,文中方法更具优势,1.4s求解105样本规模的ijcnn文本库.
Support Vector Data Description (SVDD) is an unsupervised learning algorithm, which has important applications in the field of image recognition and information security.Coordinate descent method is an effective method to solve large-scale classification problems, with a simple operation flow and fast convergence rate In this paper, we propose an efficient dual-coordinate descent algorithm for large-scale SVDD, which can obtain analytic solutions for sub-problems in each iteration and reduce the computational complexity by using accelerating strategies and simple operations. , And compared their advantages and disadvantages.Experiments verify the simulation and real large-scale database algorithm validation.Compared with LibSVDD, the method in this paper is more advantageous, 1.4s to solve 105 sample size ijcnn text library.