Learning Invariant Representation with Consistency and Diversity for Semi-supervised Source Hypothesis Transfer

Abstract

Semi-supervised domain adaptation (SSDA) aims to solve tasks in target domain by utilizing transferable information learned from the available source domain and a few labeled target data. However, source data is not always accessible in practical scenarios, which restricts the application of SSDA in real world circumstances. In this paper, we propose a novel task named Semi-supervised Source Hypothesis Transfer (SSHT), which performs domain adaptation based on source trained model, to generalize well in target domain with a few supervisions. In SSHT, we are facing two challenges: (1) The insufficient labeled target data may result in target features near the decision boundary, with the increased risk of mis-classification;(2) The data are usually imbalanced in source domain, so the model trained with these data is biased. The biased model is prone to categorize samples of minority categories into majority ones, resulting in low prediction diversity. To tackle the above issues, we propose Consistency and Diversity Learning (CDL), a simple but effective framework for SSHT by facilitating prediction consistency between two randomly augmented unlabeled data and maintaining the prediction diversity when adapting model to target domain. Encouraging consistency regularization brings difficulty to memorize the few labeled target data and thus enhances the generalization ability of the learned model. We further integrate Batch Nuclearnorm Maximization into our method to enhance the discriminability and diversity. Experimental results show that our method outperforms existing SSDA methods and unsupervised model adaptation methods on DomainNet, Office-Home and Office-31 datasets. The code is available at https://github.com/Wang-xd1899/SSHT.

Multi-level Consistency Learning for Semi-supervised Domain Adaptation

“…Model adaptation (MA) was proposed to implement domain adaptation without access to the source data, thereby addressing the dilemma of data sharing versus data privacy [8] in traditional domain adaptation. Many works have been proposed to address the problem of MA [8], [9], [11], [12], [25]- [36], and the methods can be roughly categorized into two streams: generative [25], [27] and discriminative [8], [9], [26]. Generative methods usually model the generation of labeled images or features.…”

“…We compared our method with i) baseline methods: only train the labeled target samples with cross entropy loss (CE), and CE with entropy minimization for unlabeled target samples (ENT) [58], ii) semi-supervised domain adaptation (SSDA) methods: minimax entropy (MME) [6], CDAC [7], iii) semi-supervised learning (SSL) methods: MixMatch [59], FixMatch [15], iv) semi-supervised model adaptation (SSMA) methods: SHOT++ [32], SSHT [12], and v) a universal model adaptation method: UMA [29]. To make a fair comparison, we implement them with the same F (•|θ s ) as used in our method.…”

Universal Semi-supervised Model Adaptation via Collaborative Consistency Training

“…Although promising, existing SSDA methods usually assume that the source data is available during training, which is impractical in many real-world scenarios where restrictions apply, e.g., data privacy and limited storage [28]. To meet such new demands, a new research topic, namely the model adaptation [28,32,33,50,57], has recently been proposed with the aim of transferring knowledge from a pretrained source model rather than the source data. To simplify the problem, most of these works assume that the source and target domain share the same label set.…”