基于加权分类损失和核范数的领域自适应模型

领域自适应将源域上学习到的知识迁移到目标域上，使得在带标签数据少的情况下也可以有效地训练模型。采用伪标签的领域自适应模型未考虑错误伪标签的影响，并且在决策边界处样本的分类准确率较低，针对上述问题提出了基于加权分类损失和核范数的领域自适应模型。该模型使用带有伪标签的可信样本特征与带有真实标签的源域样本特征构建辅助域，在辅助域上设计加权分类损失函数，降低错误伪标签在训练过程中产生的影响；加入批量核范数最大化损失，提高决策边界处样本的分类准确率。在Office31、Office-Home、Image-CLEFDA基准数据集上与之前模型的对比实验表明，该模型有更高的精确度。

Domain adaptation based on weighted classification loss and nuclear-norm

Domain adaptation transfers the knowledge learned from the source domain to the target domain, so that the model can be effectively trained in the case of less labeled data. The domain adaptation models using pseudo-labels do not consider the influence of false pseudo-labels, and the classification accuracy of samples at the decision boundary is low. For the above problems, this paper proposed a domain adaptation model based on weighted classification loss and nuclear-norm. The model used confident sample features and their pseudo-labels, and constructed an auxiliary domain with the source domain sample features with real labels. It designed a weighted classification loss function on the auxiliary domain reduced the influence of false labels in the training process. Batch nuclear-norm maximization loss improved the accuracy of sample pseudo-labels at the decision boundary. The comparison experiments with previous models on the benchmark datasets of Office31, Office-Home and Image-CLEFDA illustrate that this method has higher accuracy.