论文部分内容阅读
基分类器的差异性对于集成学习来说至关重要,从直观上讲,对约束重采样有潜力获得比对样本重采样更好的多样性.文中在典型相关分析算法基础上,通过引入成对约束作为监督信息对样本进行特征抽取从而形成新的训练数据.算法中集成学习的思想主要体现在成对约束的选取上,对约束进行随机重采样以获得具有多样性的基分类器.在多特征手写体数据集以及人脸数据集(Yale和AR)上进行实验考察该算法随选取的约束比例变化的情况,结果表明该方法获得比传统集成学习方法更好的性能.
The diversity of base classifiers is crucial for integrated learning, and intuitively, it has the potential to obtain better diversity than constrained resampling.Based on the classical correlation analysis algorithm, The constraint is taken as supervisory information to extract the features of the samples to form new training data.The idea of integrated learning in algorithm is mainly reflected in the selection of pairs of constraints and the random resampling of constraints to obtain a diversity base classifier. Multi-feature handwritten datasets and face data sets (Yale and AR). The results show that the proposed method achieves better performance than traditional integrated learning methods.