Backpropagation-based Parameter Tuning and Granular Alignment for Multi-target Domain Adaptation

Domain adaptation has emerged as a crucial technique for transferring knowledge between different data distributions. While significant progress has been made in adapting between single source and target domains, the challenging problem of adapting to multiple target domains simultaneously has received limited attention. This paper addresses this gap by introducing a novel framework that effectively handles multiple target domains through two key innovations: backpropagation-based parameter tuning for key features (BPTF) and progressive granular category-level alignment (PGCA). Our approach uniquely combines adaptive feature weighting with dynamic domain alignment to achieve superior performance across multiple target domains simultaneously. Firstly, we introduce the backpropagation-based parameter tuning for key features (GWCF) technique, which attenuates the detrimental impact of noisy samples and significant distribution discrepancies by prioritizing essential features. Secondly, we present the progressive granular category-level alignment (PGCA) strategy, leveraging Kullback-Leibler (KL) divergence to regulate a gradual adaptation process. This method enables a smooth transition from broad categorizations to precise alignments, thereby improving consistency across the source and multiple target domains. Extensive experimental evaluations on four benchmark datasets underscore the effectiveness of our proposed approach, consistently outperforming state-of-the-art methods in multi-target domain adaptation scenarios.