Unwanted samples from private source categories in the learning objective of a partial domain adaptation setup can lead to negative transfer and reduce classification performance. Existing methods, such as re-weighting or aggregating target predictions, are vulnerable to this issue, especially during initial training stages, and do not adequately address overlapping categorical distributions. We propose a solution to overcome these limitations by exploring beyond the first-order moments for robust alignment of categorical distributions. We employ objectives that optimize the intra and inter-class distributions in a domain-invariant fashion and design a robust pseudo-labeling for efficient target supervision. Our approach incorporates a complement entropy objective module to reduce classification uncertainty and flatten incorrect category predictions. The experimental findings and ablation analysis of the proposed modules demonstrate the superior performance of our proposed model compared to benchmarks.

翻译：在部分域适应设置的学习目标中，来自私有源类别的非目标样本可能导致负迁移并降低分类性能。现有方法（如重新加权或聚合目标预测）易受此问题影响（尤其在初始训练阶段），且未能充分处理重叠的类别分布。我们提出了一种解决方案，通过超越一阶矩探索类别分布的鲁棒对齐来克服这些限制。我们采用优化类内与类间分布以实现域不变性的目标函数，并设计了一种高效的伪标签方法用于目标监督。我们的方法引入补熵目标模块以降低分类不确定性并压制错误类别预测。实验结果表明，所提模块及其消融分析验证了模型相较于基准方法的优越性能。