去噪空间上的多核学习

Multiple kernel learning in denoising space

回顾了一种多核学习(multiple kernel learning,MKL)方法——lp范数约束的多核Fisher判别分析(lpregu-larized multiple kernel Fisher discriminant analysis,MK-FDA),研究了固定范数和p范数下MKL的性能对比,并针对原始特征空间必然存在噪点的现象,对在特征空间去噪之后的MKL方法的效果进行了探索。在VOC 2007数据集上的实验结果表明,lpMK-FDA无论使用原始核函数或者去噪后核函数的性能都超越了固定范数约束下的对比方法;特征空间的去噪处理能提高单核FDA方法和lpMK-FDA方法的性能;训练得到的核函数的权重与去噪空间中保留的特征数量存在一种正相关性。

A multiple kernel learning（MKL） technique called lp regularized multiple kernel Fisher discriminant analysis（lp MK-FDA） was reviewed,and MKL′s performance was compared fixed-norm and p-norm.According to the phenomenon that original feature space noises exist,the effect of feature space denoising on MKL was investigated.Experiments on the VOC 2007 dataset show that with both the original kernels or denoised kernels,lp MK-FDA outperforms its fixed-norm counterparts,and the feature space denoising boosts the performance of both single kernel FDA and lp MK-FDA,and also there is a positive correlation between the learnt kernel weights and the amount of variance kept by feature space denoising.