Enhanced Transport Distance for Unsupervised Domain Adaptation

Li, Mengxue; Zhai, Yiming; Luo, You-Wei; Ge, Pengfei; Ren, Chuan-Xian

doi:10.1109/cvpr42600.2020.01395

Cited by 163 publications

(63 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The classification accuracy in Table 1 is reported after 100 epochs of training. The state-of-the-art methods we used as comparison in supervised setting(S) are CCSA [2], d-SNE [7] and FADA [8], we also include several state-ofthe-art unsupervised methods(U) [5,6,4] for comparison. Cross-entropy (CE) classification for source and target data are also compared as a baseline.…”

Section: Methodsmentioning

confidence: 99%

“…The typical solution is to use another available dataset on a closely related task, which leads to the problem of domain adaptation [1]. Existing domain adaptation methods can be supervised [2], semi-supervised [3] or unsupervised [4,5,6]. The main task is to learn the knowledge from the source domain, which is adapted to the target domain.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Hardmix: A Regularization Method to Mitigate the Large Shift in Few-Shot Domain Adaptation

Liang

Yang

2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Few-Shot Domain Adaptation aims to transfer knowledge learned from a known domain to a closely related novel domain with only a few training data available for each class. The limited number of target training data makes it challenging to bridge the domain gap and can easily lead to overfitting. In this paper, we proposed HardMix as a regularization technique which interpolates the data in feature space and assigns augmented features with 'hard' labels to eliminate the domain discrepancy. In order to generate a better decision boundary and a more compact intra-class distribution, an adaptive triplet loss is proposed to constrain the 'hard' samples near the decision boundary. We demonstrated its effectiveness by comparing our results with the state-of-the-art methods on several benchmark datasets.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Hardmix: A Regularization Method to Mitigate the Large Shift in Few-Shot Domain Adaptation

Liang

Yang

2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

show abstract

“…Adversarial based methods [9,33] alternatively optimize the feature generator and domain discriminator, which are respectively supposed to be domain-confusable and discriminative, to achieve domain confusion. Extended from the marginal distribution assumption, recent works [4,6,20,22,23] show that the models yield promising results by introducing the label information. Joint Adaptation Network (JAN) [23] builds a joint distribution alignment model via the features from different hidden layers.…”

Section: Related Workmentioning

confidence: 99%

“…Ben-David et al [2] give a theoretical insight into the domain adaptation problem, they show that the risk of the target domain is mainly bounded by the risk of the source domain and the discrepancy between distributions of two domains. Inspired by this theory, many methods are proposed to mitigate the discrepancy between feature distributions of the source and target domains, e.g., explicit discrepancy minimization via Maximum Mean Discrepancy (MMD) [13,21], domain invariant feature learning [26], Optimal Transport (OT) based feature matching [7,20,37], manifold based feature alignment [10], statistical moment matching [21,32] and adversarial domain adaptation [9]. These methods are proved to be effective in minimizing the marginal discrepancy and alleviating the domain shift problem.…”

Section: Introductionmentioning

confidence: 99%

Conditional Bures Metric for Domain Adaptation

Luo¹,

Ren²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

As a vital problem in classification-oriented transfer, unsupervised domain adaptation (UDA) has attracted widespread attention in recent years. Previous UDA methods assume the marginal distributions of different domains are shifted while ignoring the discriminant information in the label distributions. This leads to classification performance degeneration in real applications. In this work, we focus on the conditional distribution shift problem which is of great concern to current conditional invariant models. We aim to seek a kernel covariance embedding for conditional distribution which remains yet unexplored. Theoretically, we propose the Conditional Kernel Bures (CKB) metric for characterizing conditional distribution discrepancy, and derive an empirical estimation for the CKB metric without introducing the implicit kernel feature map. It provides an interpretable approach to understand the knowledge transfer mechanism. The established consistency theory of the empirical estimation provides a theoretical guarantee for convergence. A conditional distribution matching network is proposed to learn the conditional invariant and discriminative features for UDA. Extensive experiments and analysis show the superiority of our proposed model.

show abstract

“…We compare GET with several SOTA UDA methods: DAN (Long et al, 2015), DANN (Ganin & Lempitsky, 2015), SAFN (Xu et al, 2019), ETD (Li et al, 2020), ALDA (Chen et al, 2020), DMP (Luo et al, 2020), BNM (Cui et al, 2020), DWL (Xiao & Zhang, 2021).…”

Section: Comparisonmentioning

confidence: 99%

Discriminative and Geometry-Aware Unsupervised Domain Adaptation

Luo

Chen

et al. 2020

IEEE Trans. Cybern.

View full text Add to dashboard Cite

Domain adaptation (DA) aims to generalize a learning model across training and testing data despite the mismatch of their data distributions. In light of a theoretical estimation of upper error bound, we argue in this paper that an effective DA method should 1) search a shared feature subspace where source and target data are not only aligned in terms of distributions as most state of the art DA methods do, but also discriminative in that instances of different classes are well separated; 2) account for the geometric structure of the underlying data manifold when inferring data labels on the target domain. In comparison with a baseline DA method which only cares about data distribution alignment between source and target, we derive three different DA models, namely CDDA, GA-DA, and DGA-DA, to highlight the contribution of Close yet Discriminative DA(CDDA) based on 1), Geometry Aware DA (GA-DA) based on 2), and finally Discriminative and Geometry Aware DA (DGA-DA) implementing jointly 1) and 2). Using both synthetic and real data, we show the effectiveness of the proposed approach which consistently outperforms state of the art DA methods over 36 image classification DA tasks through 6 popular benchmarks. We further carry out in-depth analysis of the proposed DA method in quantifying the contribution of each term of our DA model and provide insights into the proposed DA methods in visualizing both real and synthetic data.

show abstract

Enhanced Transport Distance for Unsupervised Domain Adaptation

Cited by 163 publications

References 4 publications

Hardmix: A Regularization Method to Mitigate the Large Shift in Few-Shot Domain Adaptation

Hardmix: A Regularization Method to Mitigate the Large Shift in Few-Shot Domain Adaptation

Conditional Bures Metric for Domain Adaptation

Discriminative and Geometry-Aware Unsupervised Domain Adaptation

Contact Info

Product

Resources

About