基于双重权重偏差建模的无监督域适应

Unsupervised Domain Adaptation Based on Weighting Dual Biases

无监督域适应(Unsupervised Domain Adaptation,UDA)是一类新兴的机器学习范式,其通过对源域知识在无标记目标域上的迁移利用,来促进目标域模型的训练。为建模源域与目标域之间的域分布差异,最大均值差异(Maximum Mean Discrepancy,MMD)建模被广泛应用,其对UDA的性能提升起到了有效的促进作用。然而,这些方法通常忽视了领域之间对应类规模与类分布等结构信息,因为目标域与源域的数据类规模与数据分布通常并非一致。为此,文中提出了一种基于跨域类和数据样本双重加权的无监督域适应模型(Sample weighted and Class weighted based Unsupervised Domain Adaptation Network,SCUDAN)。具体而言,一方面,通过源域类层面的适应性加权来调整源域类权重,以实现源域与目标域之间的类分布对齐;另一方面,通过目标域样本层面的适应性加权来调整目标域样本权重,以实现目标域与源域类中心的对齐。此外,文中还提出了一种CEM(Classification Expectation Maximization)优化算法,以实现对SCUDAN的优化求解。最后,通过对比实验和分析,验证了所提模型和算法的有效性。

Unsupervised domain adaptation(UDA)is a new kind of machine learning paradigm,which facilitates the training of target domain model through transferring knowledge from source domain to unlabeled target domain.In order to model the domain distribution difference between the source domain and target domain,the maximum mean discrepancy(MMD)is widely applied,it plays an effective role in promoting the performance of UDA.Usually,the class size and data distribution between the target domain and the source domain are not the same,unfortunately,these methods usually ignore this structure information.To this end,this paper proposes a model called sample weighted and class weighted based unsupervised domain adaptation network(SCUDAN).On one hand,the class distribution alignment between the source domain and the target domain is achieved through adaptive weighting on the classes of the source domain.On the other hand,the class centers between the target domain and the source domain can be aligned through adaptive weighting on the samples of the target domain.In addition,a CEM(Classification Expectation Maximization)algorithm is proposed to optimize SCUDAN.Finally,the effectiveness of the proposed method is verified by comparative experiments and analysis.