Unsupervised domain adaptation: A multi-task learning-based method

Abstract

This paper presents a novel multi-task learningbased method for unsupervised domain adaptation. Specifically, the source and target domain classifiers are jointly learned by considering the geometry of target domain and the divergence between the source and target domains based on the concept of multi-task learning. Two novel algorithms are proposed upon the method using Regularized Least Squares and Support Vector Machines respectively. Experiments on both synthetic and real world cross domain recognition tasks have shown that the proposed methods outperform several state-of-the-art domain adaptation methods.

Multi-Adversarial Partial Transfer Learning With Object-Level Attention Mechanism for Unsupervised Remote Sensing Scene Classification

“…The trend is similar to the results listed in Tables 2 and 3. mtUDA [47] outperforms ADDA [3] with 0.4% higher. The partial transfer methods like SAN [7], WAN [62] and our MOAN are designed with special GAN-based structures to evaluate the samples and features, therefore showing obvious advantages in processing these partial transfer problems.…”

“…Recent adversarial learning-based methods (like ADDA [3]) have higher accuracies than classical methods. The multi-task method, mtUDA [47], performs better than ADDA [3] on both 2 tasks. Those methods proposed for partial transfer learning, SAN and WAN [62], outperform other existing methods.…”

“…There are always multiple sources, more than 2, in those applications of them [43]. Reference [47] did not adopt adversarial learning-based structure and the classifications in source and target domains were regarded as 2 tasks. These 2 classifiers were boosted by a joint learning and regularized by Maximum Mean Discrepancy (MMD).…”

“…In addition to adversarial learning based methods above, we also adopt the novel multi-task method mtUDA [47] for comparison. Self-supervised learning method AMDIM [66] is also used in our experiments.…”

Recent Advances in Transfer Learning for Cross-Dataset Visual Recognition

“…While the drawbacks are the computational cost and the local optimum. To date, these two approaches have produced similar performance on some datasets [116,152,162,274] though the end-to-end deep systems involve more parameters and require more computational costs. One of the missing study in the literature is a systematic study and comparison of the two approaches under same or similar conditions.…”