如何克服噪声影响是数据挖掘中的经典问题。随着当前数据量越来越大，噪声的数量和种类也越来越多，噪声对模型的影响越来越甚，这一问题也越来越突出。传统的支持向量机使用铰链损失函数，会放大噪声的影响。在支持向量机的研究中，如何降低噪声的影响已经成为热点问题。因此，本项目拟引进不同的非凸损失函数处理噪声数据，其中特别注重利用非凸优化技术提取噪声数据中本质的有价值信息。具体研究内容包括：1、对于噪声数据的分类问题，引进新的非光滑有界非凸损失函数，从而提出适应于不同噪声的Ramp-类型损失函数支持向量机模型；2、对于噪声数据的回归问题和半监督问题，构造光滑非凸损失函数支持向量机模型；3、利用稀疏Lp-模非凸损失函数支持向量机研究噪声数据的特征选择问题；4、研究处理噪声数据非凸损失函数支持向量机模型的快速求解算法。目前对噪声数据的研究，不仅是热点，也是难点。因此，从理论和实践方面来说本项目都具有创新意义。

It is a classic problem to overcome the influence of noise in data mining. With the increasing amount of data, there are more and more numbers and types of noise. The impact of noise on the model has become increasingly severe, and this problem has become more and more prominent. The traditional support vector machine (SVM) using the hinge loss function, which will amplify the effect of noise on the model. How to reduce the impact of noise on SVM is a hot issue. Therefore, this project intends to introduce different non-convex loss functions to deal with noise data, with particular emphasis on the use of non-convex optimization techniques to extract valuable information inherent in noise data. The specific research contents include: 1. For classification problems with noise data, introduce new non-smooth non-convex bounded loss functions, such as Ramp-loss type functions in support vector machine model so as to adapt to different noises; 2. For regression and semi-supervised problems with noise data, construct the smooth non-convex loss function support vector machine models; 3. For the feature selection problem of noise data study the sparse Lp-norm non-convex loss function support vector machines; 4. Study the fast algorithms for non-convex loss function support vector machine with noise data. At present, the study of the noise data is a hot and difficult issue. Therefore, the project is of innovative significance in both theory and practice.