基于最大分布加权均值嵌入的领域适应学习

最大均值差异仅用于反映样本空间总体的分布信息和全局结构信息，忽略了单个样本对全局度量贡献的差异性。为此，提出一种最大分布加权均值差异（MDWMD）度量方法，采用白化余弦相似性度量为源域和目标域的所有样本设计相应的分布权重，使得每个样本的分布差异信息在全局度量中均得以体现。进一步，在MDWMD基础上，结合联合分布调整思想，提出一种领域适应学习算法：基于最大分布加权均值嵌入的联合分布调整，同时对源域和目标域中的数据进行边缘概率分布调整和条件分布调整。实验结果表明，与现有典型的迁移学习和无迁移学习算法相比，所提算法在不同类型跨领域图片数据集上的分类精度较高。

Domain Adaptation Learning Based on Maximum Distribution Weighted Mean Discrepancy

Maximum mean discrepancy is merely used to reflect the overall distribution information and the global structure information of sample space, which neglects the difference of contribution of each sample to the global measure. In this paper, a kind of maximum distribution weighted mean discrepancy (MDWMD) is proposed where the whitened cosine similarity is used to design distribution weights for all samples from the source and the target domains. Thus, the distribution discrepancy information of each sample can be reflected in the global measure. Further, based on the MDWMD measure and the idea of joint distribution adaptation (JDA), a kind of domain adaptation learning algorighm called JDA-MDWMD is proposed. The JDA-MDWMD can simultaneously adjust the marginal probability distribution and the conditional distribution of the source and the target domains. Experimental results show that, compared with the typical transfer learning and non-transfer learning algorithms, the JDA-MDWMD yields higher classification accuracy on different types of cross-domain image datasets.