Multi-View Collaborative Learning for Semi-Supervised Domain Adaptation

Abstract

Recently, Semi-supervised Domain Adaptation (SSDA) has become more practical because a small number of labeled target samples can significantly boost the empirical target performance when using SSDA. Several current methods focus on prototype-based alignment to achieve cross-domain invariance in which the labeled samples from the source and target domains are concatenated to estimate the prototypes. The model is then trained to assign the unlabeled target data to the prototype within the same class. However, such methods fail to exploit the advantage of using few labeled target data because the labeled source data dominate the prototypes in the supervision process. Moreover, a recent method [1] showed that concatenating source and target samples for training can damage the semantic information of representations, which degrades the trained model's ability to generate discriminative features. To solve these problems, in this paper, we divide labeled source and target samples into two subgroups for training. One group includes a large number of labeled source samples, and the other obtains a few labeled target samples. Then, we propose a novel SSDA framework that consists of two models. A model trained on the group that has the labeled source samples to provide an "inter-view" on the unlabeled target data is called the inter-view model. A model trained on a few labeled target samples that provides an "intra-view" of the unlabeled target data is called the intra-view model. Finally, both of these models collaborate to fully exploit information on the unlabeled target data. To the best of our knowledge, our proposed method achieves the state-of-the-art classification performance of SSDA in extensive experiments conducted on several visual benchmark domain adaptation datasets that utilize the advantages of multiple views and collaborative training.

Dual Dynamic Consistency Regularization for Semi-Supervised Domain Adaptation

“…Specifically, [1], [11], [13], [15], [29]- [31], [33], [39] utilized a combination of a CNN encoder followed by the MLP classifier in both stages. To alleviate bias confirmation, [9], [12] introduced a new framework by adding one more MLP classifier, which consists of a single CNN encoder and two MLP classifiers. [14] was the first SSDA approach that used two CNN encoders and two MLP classifiers with the co-training strategy to boost the classification accuracy.…”

“…The semi-supervised domain adaptation scenario, as depicted in [9]- [18], is extensively employed to yield remarkable classification accuracy compared to the unsupervised domain adaptation setting [19]- [23]. This is because a model trained under UDA is only accessed to labeled source data, while a model trained with the SSDA setting benefits from the extra target information with a few labeled target samples besides labeled source data.…”

“…As a result, the model often prioritizes the abundant label information from source data, potentially paying less attention to target label information and introducing bias in learning. To address this problem, [14] and [9] divide SSDA into two different tasks, such as UDA [24], [25]-(labeled source+unlabeled target) and semi-supervised learning (SSL) [27], [28]-(labeled target+unlabeled target). Each task con-sists of the different labeled sets used to train the different models.…”

“…Then, they exchange their knowledge through the cotraining strategy to make consistency on the unlabeled target data. Many SSDA methods [9]- [13], [15], [17], [18] follow a similar methodology in designing network architecture for data representation extraction. They employ convolutional neural networks (CNNs) as encoders, followed by multilayer perceptrons (MLPs) as classifiers.…”

“…Furthermore, SSDA [9], [11], [33] methods show outstanding classification results by exploiting the unlabeled target information. The core technique of these methods is to combine consistency regularization [28] and pseudo-labeling [38].…”