基于稀疏贝叶斯的流形学习

Manifold Learning Based on Sparse Bayesian Approach

针对当前监督学习算法在流形数据集上分类性能的缺陷,如分类精度低且稀疏性有限,本文在稀疏贝叶斯方法和流行正则化框架的基础上,提出一种稀疏流形学习算法(Manifold Learning Based on Sparse Bayesian Approach,MLSBA).该算法是对稀疏贝叶斯模型的扩展,通过在模型的权值上定义稀疏流形先验,有效利用了样本数据的流形信息,提高了算法的分类准确率.在多种数据集上进行实验,结果表明:MLSBA不仅在流形数据集上取得良好的分类性能,而且在非流形数据集上效果也比较好;同时算法在两类数据集上均具有良好的稀疏性能.

Aiming at the classification performance deficiencies of current supervised learning algorithms on manifold data sets,e. g. low classification accuracy and limited sparsity,a sparse manifold learning algorithm based on sparse Bayesian inference and manifold regularization framework is proposed. The algorithm is called manifold learning based on sparse Bayesian approach（ MLSBA）. MLSBA is an extension of sparse Bayesian model,by introducing sparse manifold priors to the weights,which can effectively employ the manifold information of sample data to improve the classification accuracy.Extensive experiments are conducted on various datasets, and the results show that MLSBA not only achieves better classification performance on manifold datasets,but also has comparable effectiveness on the non-manifold datasets, and our algorithm has good sparsity on two categories of datasets at the same time.