国家自然科学基金(61170151); 江苏省自然科学基金(BK2011728); 江苏省“青蓝”工程创新团队项目
多视图学习方法通过视图间互补信息的融合,达到增强单一视图方法的鲁棒性并提升学习性能的目的.典型相关分析(canonical correlation analysis,简称CCA)是一种重要的多视图信息融合技术.其研究的是针对同一组目标两组不同观测数据间的相关性,目标是得到一组相关性最大的投影向量.但当面对标号有序的分类任务时,CCA因没有利用类信息和类间有序信息,造成了对分类性能的制约.为此,通过将有序类信息嵌入CCA进行扩展,发展出有序判别典型相关分析(ordinal discriminative canonical correlation analysis,简称OR-DisCCA).实验结果表明, OR-DisCCA的性能比相关方法更优.
Multi-View learning is a method to improve the robustness and learning performance of single-view learning by fusing the complementary information. Canonical correlation analysis (CCA) which is used to analyze correlation between two datasets of the same objects is an important method for multi-view feature fusion. CCA aims to seek a pair of projections associated with the two sets of data such that they are maximally correlated. However, CCA results in constraint of the classification performance due to not utilizing the class information or ordinal information of different classes for some applications in which the data labels are ordinal. In order to compensate such a shortcoming, ordinal discriminative canonical correlation analysis (OR-DisCCA) is proposed in this paper by incorporating the class information and ordinal information for extending the traditional CCA. The experimental results show that OR-DisCCA outperforms existing related methods.