Pith. sign in

REVIEW 3 major objections 5 minor 86 references

Beyond Batch Learning: Global Awareness Enhanced Domain Adaptation

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Adding hand-defined class-specific anchor templates to adversarial domain adaptation gives reported state-of-the-art accuracy on 27 unsupervised transfer tasks.

desk verdict The core PFR alignment loss in Eqs. (11)–(12) is dimensionally ill-defined under the paper's own definitions, and the reported SOTA claims are overstated; worth a look as a salvageable idea, not as a paper in its current form. read the letter →

arxiv 2502.06272 v1 pith:77CMWOYA submitted 2025-02-10 cs.LG

Lingkun Luo , Shiqiang Hu , Liming Chen This is my paper · ORCID
classification cs.LG
keywords domainadaptationunsupervisedglobalawarenesspredefinedfeaturerepresentationmaximummeandiscrepancyadversariallearningbatchimageclassification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that batch-trained deep domain adaptation models suffer from 'global ignorance': each mini-batch shows only a local slice of the data, so cross-domain alignment can be asymmetric or self-contradictory. The proposed fix, GAN-DA, adds a predefined feature representation (PFR) loss that pulls source and target sub-domains toward fixed class-wise anchor vectors, with an orthogonal component that mimics the one-hot label space and a common component that preserves shared structure. Combined with conditional adversarial learning, the method reports state-of-the-art accuracy on 27 unsupervised domain adaptation tasks (Office-31 90.2%, Office-Home 70.6%, ImageCLEF-DA 89.4%). If the claim is right, then global statistical and geometric awareness can be injected into deep DA without abandoning batch training, simply by adding a fixed prior.

What carries the argument

The load-bearing object is the predefined feature representation (PFR): a fixed, class-wise template living in the same $D$-dimensional space as the generator's output. In Eq. (12) the c-th anchor is $(1/\alpha_{epoch}) f^c_{ofr} \oplus \alpha_{epoch} f^c_{cfr}$, where $f^c_{ofr}$ is a $C \cdot m$ block vector with $m = \lfloor D/(C+1)\rfloor$ and $f^c_{cfr}$ is a length-$n = D - c m$ all-ones vector. The MMD terms in Eq. (12) measure the distance from batch-sampled sub-domain features to these anchors, giving every batch a global reference point that does not depend on which samples happen to appear; the OFR/CFR scaling provides a curriculum from shared structure to discriminative structure, and the multilinear-conditioned discriminator then merges the aligned features with labels.

What would settle it

Run a dimension check on the two arguments of the MMD norm in Eq. (12): with $m=\lfloor D/(C+1)\rfloor$ and $n=D-cm$, the vector $f^c_{ofr} \oplus f^c_{cfr}$ has length $D+(C-c)m$ for class $c$, which equals $D$ only when $c=C$; any implementation that computes the norm must either project or otherwise change the stated template, exposing the discrepancy. A corrected re-implementation could then compare GAN-DA with and without the PFR loss on a single benchmark such as Office-Home.

Watch

Extended reading notes

Core claim

The central discovery is that the 'global ignorance' problem of batch learning can be addressed with a hand-coded prior over feature geometry. GAN-DA defines a predefined feature representation $F_{pfr} = (f^c_{ofr} \oplus f^c_{cfr})_{c=1}^C$: for class $c$, $f^c_{ofr}$ is an orthogonal block vector (ones in its own block, zeros elsewhere) that imitates the one-hot label space, and $f^c_{cfr}$ is an all-ones vector capturing commonality across sub-domains. The training objective adds MMD alignment between batch-sampled source and target sub-domains and these fixed anchors, scaled by $\alpha = 1/(C+1)$ so that OFR is emphasized and CFR is de-emphasized over epochs. The paper reports that this on top of adversarial domain alignment outperforms 24 existing methods across 27 tasks and, in ablations, converges in 3–5 epochs where the base CDAN model needs about 8.

Load-bearing premise

The method's alignment loss assumes that the hand-built class template has the same number of dimensions as the feature vector the network produces, but the paper's own formulas give that equality only for the last class, so the loss cannot be computed as written for most classes.

Editorial extensions

If this is right

  • The auxiliary PFR loss can be added to existing adversarial DA pipelines without changing the batch training loop, because the anchor template is fixed a priori.
  • Convergence speeds up: in the reported SVHN-to-MNIST ablation, GAN-DA reaches around 96% accuracy in 3–5 epochs while the CDAN base takes roughly 8 epochs to reach about 89%.
  • The two template components are separable and individually testable: ablations show removing the CFR term degrades accuracy, removing the OFR term collapses the model to about 10%, and fixing CFR as all-ones restores 89.6%, so the paper attributes global awareness specifically to the OFR/CFR combination.
  • The reported averages on Office-31 (90.2%), ImageCLEF-DA (89.4%), and digits (97.2%) establish the claimed state-of-the-art only if the comparison numbers from previous papers are accepted.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test this, but the anchor-template idea should generalize to partial, open-set, or multi-source DA by redefining the anchor set; it is an auxiliary loss that can sit on top of any feature extractor.
  • A concrete check of Eq. (12) with the paper's definitions shows that the concatenated anchor has length $D + (C-c)m$, which only equals $D$ for $c=C$; the reported results may therefore depend on an implementation that silently corrects or projects the dimensions.
  • On Office-Home the best shallow global baseline in the table scores 72.1% against GAN-DA's 70.6%, which suggests the claimed global-awareness advantage is not universal and may fade exactly where a non-batch global method can be run.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes GAN-DA, an unsupervised domain adaptation method that augments a CDAN-style adversarial framework with a hand-defined 'predefined feature representation' (PFR) loss. The PFR is decomposed into an orthogonal feature representation (OFR) and a common feature representation (CFR), designed to encode global statistical and geometric awareness so that batch training can approximate the full-data objectives of the domain adaptation error bound. The authors report results on 27 tasks across Digits, Office-31, Office-Home, and ImageCLEF-DA, claiming state-of-the-art accuracy and including an ablation study of the proposed components.

Significance. The motivating problem is legitimate: batch-based deep domain adaptation does lose information about global distribution statistics, and the idea of anchoring batches to a fixed global representation is worth exploring. The paper also has a structured ablation that separates the OFR and CFR contributions. However, the central loss is internally inconsistent under the paper's own definitions, and the empirical validation is built on an outdated baseline pool that does not support the stated 'state-of-the-art' claim. If the loss formulation were repaired and the experiments re-run against current methods, the underlying idea could still be useful, but in its present form the contribution cannot be evaluated because the key objective is not a well-defined function.

major comments (3)
  1. [III-B3, Eqs. (11)-(12)] The concatenated PFR anchor has dimension D + (C−c)·m, not D, for every class c < C. The manuscript defines f_c_ofr as a 1×(C·m) vector with m = ⌊D/(C+1)⌋, and f_c_cfr as a 1×n vector with n = D − c·m. Their concatenation used in Eq. (11) and (12) therefore has length C·m + (D − c·m) = D + (C−c)·m. Since the generator features φ(x_i) are D-dimensional, the MMD subtraction in Eq. (12) is shape-compatible only when c = C. For Office-Home (D=2048, C=65, m=31), the c=1 anchor has length 4032, so the loss is ill-typed for 64 of 65 classes. The α/1−α scaling in Eq. (12) changes magnitudes, not dimensions, and cannot resolve the mismatch. This is not a peripheral typo: the PFR loss is the mechanism claimed to supply the global statistical/geometric awareness. The objective as written cannot be computed, so the reported results cannot be reproduced from the text. A corrected definition (e.g., n = D − C·m for all c) would make the concatenation well-defined, but the authors must also confirm that their implementation matched the corrected formula.
  2. [IV-C and IV-D] The claim of outperforming 24 established DA methods is not supported by the presented comparisons. The baseline list is dominated by methods from 2014-2020, and strong published UDA methods from 2021-2024 are absent. More directly, in the Office-Home table (Fig. 10), the authors' own GAN-DA average of 70.6% is below DOLL-DA's 72.1%, and several tasks in the same table also show GAN-DA below CDAN or GSDA. Thus the abstract's statement that GAN-DA 'outperforms 24 established DA methods by a significant margin' is contradicted by the paper's own tables. A thorough comparison against current state-of-the-art methods, with a clear statement of which methods are included and why, is essential before any superiority claim can be made.
  3. [IV-B and IV-E2] The experimental section lacks the details needed to reproduce the method even after correcting the dimension issue. The α_epoch schedule is not defined: Eq. (12) uses α_epoch but the text only says that α and 1/α are applied 'during the initial two to three epochs', with no specification of how α varies across epochs. The ablation study contains contradictory descriptions: §IV-E2 says GAN-OFR sets the OFR term to a 0 matrix, while the bullet list says 'the OFR term is initially set to 0 and the CFR term is reduced to a 1 × 10−3 matrix after Epoch 3'; these are different configurations. No code is provided, no error bars are given, and the number of random runs is not reported. Given these omissions and the central loss inconsistency, the experimental tables alone do not establish the method's effectiveness.
minor comments (5)
  1. [IV-C] In the baseline list, BSP appears twice (items 9 and 23); one of the entries is likely intended to be a different method such as TADA. Please correct the duplicate.
  2. [Figures 7-10] All four accuracy tables (Digits, Office-31, ImageCLEF-DA, Office-Home) share the caption 'COMPARISONS OF RECOGNITION RATES(%) ON OFFICE+CALTECH-256 DATABASE (SURF FEATURES)', which is wrong for most of these figures and makes the tables hard to interpret.
  3. [IV-E2] The abbreviations GAN-OFR and GAN-ORF are used inconsistently within the same subsection; please use one spelling consistently.
  4. [III-A, Eq. (2)] The MMD definition writes integrals against du_i and dv_i without explicit probability measures; it should be written in terms of expectations under U and V to be mathematically precise.
  5. [Footnote on page 1] The first-page footnote states 'This paper has been accepted for publication in IEEE TPAMI in 2025.' If that is the case, the arXiv version should be marked as such; otherwise the statement is misleading to readers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the PFR anchors are hand-specified, alpha is a fixed closed-form schedule, and the reported accuracy is measured on held-out external benchmarks.

full rationale

The paper's central claim is that adding a predefined feature representation (PFR) loss to CDAN-style adversarial adaptation improves accuracy. The anchors f_ofr and f_cfr are defined by closed-form formulas using only the number of classes C and feature dimension D, and alpha is set to 1/(C+1) as a fixed schedule. No parameter is fitted to the reported accuracies or to target labels and then recycled as a prediction. The evaluation uses held-out target-domain accuracy on Office-31, Office-Home, ImageCLEF-DA, and Digits, compared against independently reported baselines, so the claimed gains are not forced by construction. The paper's self-citations appear in related-work discussion and as a conceptual pointer for manifold unification, not as an unverified uniqueness theorem or as the load-bearing evidence for the PFR's effectiveness. The main risks are internal consistency (e.g., the dimension of f_ofr concatenated with f_cfr versus D in Eq. (12)) and experimental completeness, which are correctness concerns rather than circular reasoning.

Assumptions & free parameters 3 free parameters · 4 assumptions · 3 invented entities

The central claim rests on hand-designed PFR/OFR/CFR vectors, a hand-set alpha schedule, pseudo-label routing, and the assumption that the concatenated anchor vector has the same dimension as the deep features. The ledger below lists the most important hand-chosen and assumed elements.

free parameters (3)
  • anchor block dimension m = m = floor(D/(C+1))
    Hand-chosen structural parameter that determines OFR block size; no principled derivation or sensitivity analysis.
  • alpha weighting schedule = alpha = 1/(C+1), applied for 'initial two to three epochs'
    Closed-form but arbitrary schedule controlling OFR vs CFR emphasis; exact epoch count is vague.
  • PFR anchor vectors = 0/1 block patterns per class
    The predefined feature representation is constructed by hand, not learned; the choice of block patterns and CFR length D-c*m is ad hoc and creates dimension mismatch.
assumptions (4)
  • standard math Ben-David et al. domain adaptation error bound (Eq. 1) applies to the model and bounds target error by source risk, H-divergence, and labeling-function difference.
    Invoked at Sect. I and III-B to motivate separate optimization of Term.1-Term.3, but no adaptation-specific theorem is proved.
  • domain assumption Target sub-domain labels can be obtained from pseudo-labels generated by a base classifier.
    Needed to define D_T^(c) in Eq. (6) and to route target samples to class-specific PFR anchors; if pseudo-labels are noisy, anchor alignment can enforce wrong clusters.
  • ad hoc to paper The concatenated OFR and CFR vectors have the same dimension as the deep feature embedding.
    Required for Eq. (11)-(12) to be computable; contradicted by definitions m=floor(D/(C+1)), n=D-c*m, which give length D+(C-c)m.
  • domain assumption CDAN multilinear conditioning correctly models joint feature-label distribution.
    GAN-DA inherits CDAN's adversarial loss Eq. (13) without modification.
invented entities (3)
  • Predefined Feature Representation (PFR)
    purpose: Hand-crafted target feature vectors used as anchor points for cross-domain MMD-like alignment.
    No external falsifiable prediction; PFR exists only inside the proposed model and is not measurable outside it.
  • Orthogonal Feature Representation (OFR)
    purpose: Class-specific orthogonal blocks that mimic one-hot label space to improve separability.
    Introduced alongside PFR; no independent evidence such as predicted labels outside the method.
  • Common Feature Representation (CFR)
    purpose: All-ones block intended to capture shared structure across sub-domains and avoid negative transfer.
    Hand-defined common component; the paper provides no independent evidence for its existence or form.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Batch Learning: Global Awareness Enhanced Domain Adaptation." pith.science (2026). https://pith.science/paper/77CMWOYA

@misc{pith2026250206272,
  author       = {Pith},
  title        = {Pith review of: Beyond Batch Learning: Global Awareness Enhanced Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/77CMWOYA}},
  note         = {Machine review of arXiv:2502.06272}
}
read the original abstract

In domain adaptation (DA), the effectiveness of deep learning-based models is often constrained by batch learning strategies that fail to fully apprehend the global statistical and geometric characteristics of data distributions. Addressing this gap, we introduce 'Global Awareness Enhanced Domain Adaptation' (GAN-DA), a novel approach that transcends traditional batch-based limitations. GAN-DA integrates a unique predefined feature representation (PFR) to facilitate the alignment of cross-domain distributions, thereby achieving a comprehensive global statistical awareness. This representation is innovatively expanded to encompass orthogonal and common feature aspects, which enhances the unification of global manifold structures and refines decision boundaries for more effective DA. Our extensive experiments, encompassing 27 diverse cross-domain image classification tasks, demonstrate GAN-DA's remarkable superiority, outperforming 24 established DA methods by a significant margin. Furthermore, our in-depth analyses shed light on the decision-making processes, revealing insights into the adaptability and efficiency of GAN-DA. This approach not only addresses the limitations of existing DA methodologies but also sets a new benchmark in the realm of domain adaptation, offering broad implications for future research and applications in this field.

Figures

Figures reproduced from arXiv: 2502.06272 by the authors.

Figure 1
Figure 1. Batch learning-induced unreliable functional learning [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the proposed GAN-DA method. In Fig.2 (a), we show examples of source (in red) and target (in blue) data such as mouse, bike, and smartphone images, which have different distributions and inherent hidden data geometric structures. Different geometric shapes represent samples of different class labels. Fig.2 (b&c) depicts the aligned data distributions by using the generator and the predefined feature … view at source ↗
Figure 3
Figure 3. The learning procedure of the VAE involves approximat￾ing the distribution qϕ(z |x) optimized over finite samples with a predefined data distribution pθ(z) = N (z, 0, I) to enhance its generative capabilities. The vanilla variational autoencoder (VAE) proposes that the limited generation diversity of the autoencoder is a consequence [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: In Fig.4 (a), we can observe that the cross sub￾domain samples are collected from different batches. However, we synchronize these samples within the pre-defined feature distributions in Fig.4 (b), which helps reduce the cross-domain divergence. The objective of the op…
Figure 6
Figure 6. Figure 6: The overall optimization procedure of the proposed [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Accuracy% on Digital Image Datasets. between cross-domain features and achieve feature alignment. MADA goes a step further by optimizing the decision boundary for multiple domain discriminators, enabling the model’s sen￾sitivity to the multimodal structures present in …
Figure 8
Figure 8. Figure 8: Accuracy% on Office-31 Dataset (ResNet-50). 3) Experiments on the ImageCLEF-DA Dataset: ImageCLEF-DA contains three public datasets, i.e., Caltech￾256 (C), ImageNet ILSVRC 2012 (I), and Pascal VOC 2012 (P), with each dataset treated as a separate domain. Each domain co…
Figure 10
Figure 10. Figure 10: Accuracy% on Office-Home Dataset (ResNet-50). 1) Visulalization of the Decision Boundary Optimization: In Fig.11, we present a comparison between the baseline method CDAN and our proposed GAN-DA in two experimental sce￾narios: a clean data distribution and a noisy dat…
Figure 9
Figure 9. Figure 9: Accuracy% on ImageCLEF-DA Dataset (ResNet-50). 4) Experiments on the Office-Home Dataset: Office-Home, introduced in DAH [71], presents a challenging benchmark for the DA task, comprising four domains with 65 object categories, resulting in 12 distinct DA tasks. In Fig…
Figure 11
Figure 11. Figure 11: Figure.11.(a&b) and Figure.11.(c&d) provide visualiza￾tions of the performance of CDAN and GAN-DA on the double moon dataset under conditions of both small and large data variance, respectively. 2) Ablation Study: While GAN-DA has demonstrated im￾pressive performance …
Figure 12
Figure 12. Figure 12: The upper table presents the accuracy of the five proposed [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

86 extracted references · 74 canonical work pages

  1. [1]

    To understand deep learning we need to understand kernel learning

    Mikhail Belkin, Siyuan Ma, and Soumik Mandal. To understand deep learning we need to understand kernel learning. In International Conference on Machine Learning , pages 541–549. PMLR, 2018. 4

    2018

  2. [2]

    Laplacian eigenmaps for dimensionality reduction and data representation

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation. Neural computation, 15(6):1373–1396,

  3. [3]

    A theory of learning from different domains

    Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. A theory of learning from different domains. Machine learning, 79(1):151–175, 2010. 1

    2010

  4. [4]

    Representation learning: A review and new perspectives

    Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence , 35(8):1798–1828, 2013. 6

    2013

  5. [5]

    Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models

    Sam Bond-Taylor, Adam Leach, Yang Long, and Chris G Willcocks. Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models. IEEE transactions on pattern analysis and machine intelligence , 2021. 4

    2021

  6. [6]

    Integrating structured biological data by kernel maximum mean discrepancy

    Karsten M Borgwardt, Arthur Gretton, Malte J Rasch, Hans-Peter Kriegel, Bernhard Sch ¨olkopf, and Alex J Smola. Integrating structured biological data by kernel maximum mean discrepancy. Bioinformatics, 22(14):e49– e57, 2006. 5

    2006

  7. [7]

    Domain separation networks

    Konstantinos Bousmalis, George Trigeorgis, Nathan Silberman, Dilip Kr- ishnan, and Dumitru Erhan. Domain separation networks. In Advances in Neural Information Processing Systems , pages 343–351, 2016. 4, 8

    work page 2016

  8. [8]

    Mix and reason: Reasoning over semantic topology with data mixing for domain generalization

    Chaoqi Chen, Luyao Tang, Feng Liu, Gangming Zhao, Yue Huang, and Yizhou Yu. Mix and reason: Reasoning over semantic topology with data mixing for domain generalization. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Proces...

    work page 2022

Show all 86 references
  1. [9]

    Trans- ferability vs

    Xinyang Chen, Sinan Wang, Mingsheng Long, and Jianmin Wang. Trans- ferability vs. discriminability: Batch spectral penalization for adversarial domain adaptation. In International conference on machine learning, pages 1081–1090. PMLR, 2019. 10

    2019

  2. [10]

    Joint distribution optimal transportation for domain adaptation

    Nicolas Courty, R ´emi Flamary, Amaury Habrard, and Alain Rakotoma- monjy. Joint distribution optimal transportation for domain adaptation. In Advances in Neural Information Processing Systems , pages 3733–3742,

  3. [11]

    Optimal transport for domain adaptation

    Nicolas Courty, R ´emi Flamary, Devis Tuia, and Alain Rakotomamonjy. Optimal transport for domain adaptation. IEEE transactions on pattern analysis and machine intelligence , 39(9):1853–1865, 2017. 2

    2017

  4. [12]

    Gradually vanishing bridge for adversarial domain adaptation

    Shuhao Cui, Shuhui Wang, Junbao Zhuo, Chi Su, Qingming Huang, and Qi Tian. Gradually vanishing bridge for adversarial domain adaptation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12455–12464, 2020. 4

    2020

  5. [13]

    Learning data representations with joint diffusion models

    Kamil Deja, Tomasz Trzci ´nski, and Jakub M Tomczak. Learning data representations with joint diffusion models. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages 543–

  6. [14]

    Cluster alignment with a teacher for unsupervised domain adaptation

    Zhijie Deng, Yucen Luo, and Jun Zhu. Cluster alignment with a teacher for unsupervised domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision , pages 9944–9953, 2019. 10 JOURNAL OF LATEX 2021 14

    2019

  7. [15]

    Robust multiview data analysis through collective low-rank subspace

    Zhengming Ding and Yun Fu. Robust multiview data analysis through collective low-rank subspace. IEEE Trans. Neural Netw. Learning Syst. , 29(5):1986–1997, 2018. 3

    1986

  8. [16]

    Unsupervised visual domain adaptation using subspace alignment

    Basura Fernando, Amaury Habrard, Marc Sebban, and Tinne Tuytelaars. Unsupervised visual domain adaptation using subspace alignment. In IEEE International Conference on Computer Vision, ICCV 2013, Sydney, Australia, December 1-8, 2013 , pages 2960–2967, 2013. 3

    2013

  9. [17]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial training of neural networks. The Journal of Machine Learning Research, 17(1):2096–2030, 2016. 1, 4

    2016

  10. [18]

    Back to the source: Diffusion-driven adaptation to test- time corruption

    Jin Gao, Jialing Zhang, Xihui Liu, Trevor Darrell, Evan Shelhamer, and Dequan Wang. Back to the source: Diffusion-driven adaptation to test- time corruption. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 11786–11796, 2023. 4, 5

    2023

  11. [19]

    Deep reconstruction-classification networks for unsupervised domain adaptation

    Muhammad Ghifary, W Bastiaan Kleijn, Mengjie Zhang, David Balduzzi, and Wen Li. Deep reconstruction-classification networks for unsupervised domain adaptation. In Computer Vision–ECCV 2016: 14th European Con- ference, Amsterdam, The Netherlands, October 11–14, 2016, Proceeding...

    2016

  12. [20]

    Deep learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016. 5

    2016

  13. [21]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems , pages 2672–2680, 2014. 4, 8

    2014

  14. [22]

    A kernel method for the two-sample-problem

    Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Sch ¨olkopf, and Alex J Smola. A kernel method for the two-sample-problem. In Advances in neural information processing systems , pages 513–520, 2007. 6

    2007

  15. [23]

    A kernel two-sample test

    Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Sch ¨olkopf, and Alexander Smola. A kernel two-sample test. The Journal of Machine Learning Research, 13(1):723–773, 2012. 6

    2012

  16. [24]

    Multi-source domain adaptation with collaborative learning for semantic segmentation

    Jianzhong He, Xu Jia, Shuaijun Chen, and Jianzhuang Liu. Multi-source domain adaptation with collaborative learning for semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11008–11017, 2021. 4

    2021

  17. [25]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. arXiv preprint arXiv:1512.03385, 2015. 4, 9

    2015 arXiv

  18. [26]

    Learning an invariant hilbert space for domain adaptation

    Samitha Herath, Mehrtash Harandi, and Fatih Porikli. Learning an invariant hilbert space for domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 3845–3854, 2017. 3

    2017

  19. [27]

    CyCADA: Cycle-consistent adversarial domain adaptation

    Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei Efros, and Trevor Darrell. CyCADA: Cycle-consistent adversarial domain adaptation. In Jennifer Dy and Andreas Krause, editors, Proceedings of the 35th International Conference on Machine Le...

    1989

  20. [28]

    Unsupervised domain adaptation with label and structural consis- tency

    Cheng-An Hou, Yao-Hung Hubert Tsai, Yi-Ren Yeh, and Yu-Chiang Frank Wang. Unsupervised domain adaptation with label and structural consis- tency. IEEE Trans. Image Processing , 25(12):5552–5562, 2016. 3

    2016

  21. [29]

    Unsupervised do- main adaptation with hierarchical gradient synchronization

    Lanqing Hu, Meina Kan, Shiguang Shan, and Xilin Chen. Unsupervised do- main adaptation with hierarchical gradient synchronization. In Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition , pages 4043–4052, 2020. 10

    2020

  22. [30]

    Discriminative radial domain adaptation

    Zenan Huang, Jun Wen, Siheng Chen, Linchao Zhu, and Nenggan Zheng. Discriminative radial domain adaptation. IEEE Trans. Image Process. , 32:1419–1431, 2023. 4

    2023

  23. [31]

    Variational deep embedding: An unsupervised and generative approach to clustering

    Zhuxi Jiang, Yin Zheng, Huachun Tan, Bangsheng Tang, and Hanning Zhou. Variational deep embedding: An unsupervised and generative approach to clustering. In Carles Sierra, editor, Proceedings of the Twenty- Sixth International Joint Conference on Artificial Intelligence, IJCAI...

    2017

  24. [32]

    Adaptive component embedding for domain adaptation

    Mengmeng Jing, Jidong Zhao, Jingjing Li, Lei Zhu, Yang Yang, and Heng Tao Shen. Adaptive component embedding for domain adaptation. IEEE Transactions on Cybernetics , 2020. 3

    2020

  25. [33]

    Detecting change in data streams

    Daniel Kifer, Shai Ben-David, and Johannes Gehrke. Detecting change in data streams. In Proceedings of the Thirtieth international conference on Very large data bases-Volume 30 , pages 180–191. VLDB Endowment,

  26. [34]

    Auto-encoding variational bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 , 2013. 6

    2013 arXiv

  27. [35]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems , pages 1097–1105, 2012. 4

    2012

  28. [36]

    Prin- cipled hybrids of generative and discriminative models

    Julia A Lasserre, Christopher M Bishop, and Thomas P Minka. Prin- cipled hybrids of generative and discriminative models. In 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’06), volume 1, pages 87–94. IEEE, 2006. 4

    2006

  29. [37]

    Sliced wasserstein discrepancy for unsupervised domain adaptation

    Chen-Yu Lee, Tanmay Batra, Mohammad Haris Baig, and Daniel Ulbricht. Sliced wasserstein discrepancy for unsupervised domain adaptation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10285–10295, 2019. 10

    2019

  30. [38]

    Maximum density divergence for domain adaptation

    Jingjing Li, Erpeng Chen, Zhengming Ding, Lei Zhu, Ke Lu, and Heng Tao Shen. Maximum density divergence for domain adaptation. IEEE transactions on pattern analysis and machine intelligence , 43(11):3918– 3930, 2020. 9, 10

    2020

  31. [39]

    Discriminative transfer feature and label consistency for cross-domain image classification

    Shuang Li, Chi Harold Liu, Limin Su, Binhui Xie, Zhengming Ding, CL Philip Chen, and Dapeng Wu. Discriminative transfer feature and label consistency for cross-domain image classification. IEEE Transactions on Neural Networks and Learning Systems , 2020. 3

    2020

  32. [40]

    Unsupervised image-to- image translation networks

    Ming-Yu Liu, Thomas Breuel, and Jan Kautz. Unsupervised image-to- image translation networks. Advances in neural information processing systems, 30, 2017. 10

    2017

  33. [41]

    Coupled generative adversarial networks

    Ming-Yu Liu and Oncel Tuzel. Coupled generative adversarial networks. In Advances in neural information processing systems , pages 469–477, 2016. 10

    2016

  34. [42]

    Homol- ogous component analysis for domain adaptation

    Youfa Liu, Weiping Tu, Bo Du, Lefei Zhang, and Dacheng Tao. Homol- ogous component analysis for domain adaptation. IEEE Transactions on Image Processing, 29:1074–1089, 2019. 4

    2019

  35. [43]

    Learning transferable features with deep adaptation networks

    Mingsheng Long, Yue Cao, Jianmin Wang, and Michael I Jordan. Learning transferable features with deep adaptation networks. In ICML, pages 97– 105, 2015. 1, 4, 6, 9, 10

    2015

  36. [44]

    Conditional adversarial domain adaptation

    Mingsheng Long, Zhangjie Cao, Jianmin Wang, and Michael I Jordan. Conditional adversarial domain adaptation. Advances in neural information processing systems, 31, 2018. 6, 8, 9, 10

    2018

  37. [45]

    Adaptation regularization: A general framework for trans- fer learning

    Mingsheng Long, Jianmin Wang, Guiguang Ding, Sinno Jialin Pan, and S Yu Philip. Adaptation regularization: A general framework for trans- fer learning. IEEE Transactions on Knowledge and Data Engineering , 26(5):1076–1089, 2013. 1, 3

    2013

  38. [46]

    Transfer feature learning with joint distribution adaptation

    Mingsheng Long, Jianmin Wang, Guiguang Ding, Jiaguang Sun, and Philip S Yu. Transfer feature learning with joint distribution adaptation. In Proceedings of the IEEE International Conference on Computer Vision , pages 2200–2207, 2013. 1, 2, 6, 9

    2013

  39. [47]

    Deep transfer learning with joint adaptation networks

    Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I Jordan. Deep transfer learning with joint adaptation networks. arXiv preprint arXiv:1605.06636, 2016. 10

    2016 arXiv

  40. [48]

    Unsu- pervised domain adaptation with residual transfer networks

    Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I Jordan. Unsu- pervised domain adaptation with residual transfer networks. In Advances in Neural Information Processing Systems , pages 136–144, 2016. 10

    2016

  41. [49]

    Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I. Jordan. Deep transfer learning with joint adaptation networks. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017 , pages 2208–2217, 2017. 4, 6

    2017

  42. [50]

    An embarrassingly simple approach to visual domain adaptation

    Hao Lu, Chunhua Shen, Zhiguo Cao, Yang Xiao, and Anton van den Hengel. An embarrassingly simple approach to visual domain adaptation. IEEE Transactions on Image Processing , 2018. 3

    2018

  43. [51]

    Knowledge transfer in vision recognition: A survey

    Ying Lu, Lingkun Luo, Di Huang, Yunhong Wang, and Liming Chen. Knowledge transfer in vision recognition: A survey. ACM Comput. Surv., 53(2):37:1–37:35, 2020. 1

    2020

  44. [52]

    Attention regularized laplace graph for domain adaptation

    Lingkun Luo, Liming Chen, and Shiqiang Hu. Attention regularized laplace graph for domain adaptation. IEEE Transactions on Image Processing , 31:7322–7337, 2022. 3, 6, 7

    2022

  45. [53]

    Discriminative and geometry-aware unsupervised domain adaptation

    Lingkun Luo, Liming Chen, Shiqiang Hu, Ying Lu, and Xiaofang Wang. Discriminative and geometry-aware unsupervised domain adaptation. IEEE Transactions on Cybernetics , 2020. 1, 3

    2020

  46. [54]

    Discriminative noise robust sparse orthogonal label regression-based domain adaptation

    Lingkun Luo, Shiqiang Hu, and Liming Chen. Discriminative noise robust sparse orthogonal label regression-based domain adaptation. International Journal of Computer Vision , 132(1):161–184, 2024. 10

    2024

  47. [55]

    Robust data geometric structure aligned close yet discriminative domain adaptation

    Lingkun Luo, Xiaofang Wang, Shiqiang Hu, and Liming Chen. Robust data geometric structure aligned close yet discriminative domain adaptation. CoRR, abs/1705.08620, 2017. 3, 9

    2017 arXiv

  48. [56]

    Domain adaptation via transfer component analysis

    Sinno Jialin Pan, Ivor W Tsang, James T Kwok, and Qiang Yang. Domain adaptation via transfer component analysis. IEEE Transactions on Neural Networks, 22(2):199–210, 2011. 2, 6

    2011

  49. [57]

    A survey on transfer learning

    Sinno Jialin Pan and Qiang Yang. A survey on transfer learning. IEEE Transactions on knowledge and data engineering, 22(10):1345–1359, 2010. 1

    2010

  50. [58]

    Transferrable prototypical networks for unsupervised domain adaptation

    Yingwei Pan, Ting Yao, Yehao Li, Yu Wang, Chong-Wah Ngo, and Tao Mei. Transferrable prototypical networks for unsupervised domain adaptation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 2239–2247, 2019. 10 15

    2019

  51. [59]

    Multi- adversarial domain adaptation

    Zhongyi Pei, Zhangjie Cao, Mingsheng Long, and Jianmin Wang. Multi- adversarial domain adaptation. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018. 4, 10

    2018

  52. [60]

    Moment matching for multi-source domain adaptation

    Xingchao Peng, Qinxun Bai, Xide Xia, Zijun Huang, Kate Saenko, and Bo Wang. Moment matching for multi-source domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1406–1415, 2019. 1, 4

    2019

  53. [61]

    Maximum classifier discrepancy for unsupervised domain adaptation

    Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku, and Tatsuya Harada. Maximum classifier discrepancy for unsupervised domain adaptation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3723–3732, 2018. 10

    2018

  54. [62]

    Generate to adapt: Aligning domains using generative adver- sarial networks

    Swami Sankaranarayanan, Yogesh Balaji, Carlos D Castillo, and Rama Chellappa. Generate to adapt: Aligning domains using generative adver- sarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 8503–8512, 2018. 10

    2018

  55. [63]

    Generalized transfer subspace learning through low-rank constraint

    Ming Shao, Dmitry Kit, and Yun Fu. Generalized transfer subspace learning through low-rank constraint. International Journal of Computer Vision , 109(1-2):74–93, 2014. 3

    2014

  56. [64]

    S. Si, D. Tao, and B. Geng. Bregman divergence-based regularization for transfer subspace learning. IEEE Transactions on Knowledge and Data Engineering, 22(7):929–942, July 2010. 2

    2010

  57. [65]

    Return of frustratingly easy domain adaptation

    Baochen Sun, Jiashi Feng, and Kate Saenko. Return of frustratingly easy domain adaptation. In AAAI, volume 6, page 8, 2016. 3

    2016

  58. [66]

    Deep coral: Correlation alignment for deep domain adaptation

    Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. In European Conference on Computer Vision , pages 443–450. Springer, 2016. 4, 10

    2016

  59. [67]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. In Computer Vision and Pattern Recog- nition (CVPR), volume 1, page 4, 2017. 1, 4, 10

    2017

  60. [68]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , July 2017. 6, 8

    2017

  61. [69]

    Deep domain confusion: Maximizing for domain invariance

    Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep domain confusion: Maximizing for domain invariance. CoRR, abs/1412.3474, 2014. 10

    2014 arXiv

  62. [70]

    The nature of statistical learning theory

    Vladimir Vapnik. The nature of statistical learning theory. Springer science & business media, 1999. 5, 6

    1999

  63. [71]

    Deep hashing network for unsupervised domain adaptation

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethu- raman Panchanathan. Deep hashing network for unsupervised domain adaptation. arXiv preprint arXiv:1706.07522 , 2017. 9, 11

    2017 arXiv

  64. [72]

    Visual domain adaptation with manifold embedded distri- bution alignment

    Jindong Wang, Wenjie Feng, Yiqiang Chen, Han Yu, Meiyu Huang, and Philip S Yu. Visual domain adaptation with manifold embedded distri- bution alignment. In 2018 ACM Multimedia Conference on Multimedia Conference, pages 402–410. ACM, 2018. 1, 3

    2018

  65. [73]

    Transferable attention for domain adaptation

    Ximei Wang, Liang Li, Weirui Ye, Mingsheng Long, and Jianmin Wang. Transferable attention for domain adaptation. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 33, pages 5345–5352, 2019. 10

    2019

  66. [74]

    Metaalign: Coordinating domain alignment and classification for unsupervised domain adaptation

    Guoqiang Wei, Cuiling Lan, Wenjun Zeng, and Zhibo Chen. Metaalign: Coordinating domain alignment and classification for unsupervised domain adaptation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16643–16653, 2021. 5

    2021

  67. [75]

    Learning semantic representations for unsupervised domain adaptation

    Shaoan Xie, Zibin Zheng, Liang Chen, and Chuan Chen. Learning semantic representations for unsupervised domain adaptation. In International conference on machine learning , pages 5423–5432. PMLR, 2018. 10

    2018

  68. [76]

    Discrim- inative transfer subspace learning via low-rank and sparse representation

    Yong Xu, Xiaozhao Fang, Jian Wu, Xuelong Li, and David Zhang. Discrim- inative transfer subspace learning via low-rank and sparse representation. IEEE Trans. Image Processing , 25(2):850–863, 2016. 3, 9

    2016

  69. [77]

    Domain-indexing variational bayes: Interpretable domain index for domain adaptation

    Zihao Xu, Guang-Yuan Hao, Hao He, and Hao Wang. Domain-indexing variational bayes: Interpretable domain index for domain adaptation. In The Eleventh International Conference on Learning Representations , 2023. 4

    2023

  70. [78]

    Mind the class weight bias: Weighted maximum mean discrepancy for unsupervised domain adaptation

    Hongliang Yan, Yukang Ding, Peihua Li, Qilong Wang, Yong Xu, and Wangmeng Zuo. Mind the class weight bias: Weighted maximum mean discrepancy for unsupervised domain adaptation. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , July 2017. 6

    2017

  71. [79]

    A survey on deep semi-supervised learning

    Xiangli Yang, Zixing Song, Irwin King, and Zenglin Xu. A survey on deep semi-supervised learning. IEEE Transactions on Knowledge and Data Engineering, 2022. 4

    2022

  72. [80]

    Fda: Fourier domain adaptation for semantic segmentation

    Yanchao Yang and Stefano Soatto. Fda: Fourier domain adaptation for semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4085–4095, 2020. 4, 8

    2020

  73. [81]

    Joint geometrical and statistical alignment for visual domain adaptation

    Jing Zhang, Wanqing Li, and Philip Ogunbona. Joint geometrical and statistical alignment for visual domain adaptation. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , July 2017. 3

    2017

  74. [82]

    Domain adaptation as a problem of inference on graphical models

    Kun Zhang, Mingming Gong, Petar Stojanov, Biwei Huang, Qingsong Liu, and Clark Glymour. Domain adaptation as a problem of inference on graphical models. Advances in neural information processing systems , 33:4965–4976, 2020. 4

    2020

  75. [83]

    Manifold criterion guided transfer learning via intermediate domain generation

    Lei Zhang, Shanshan Wang, Guang-Bin Huang, Wangmeng Zuo, Jian Yang, and David Zhang. Manifold criterion guided transfer learning via intermediate domain generation. IEEE Transactions on Neural Networks and Learning Systems , 30(12):3759–3773, 2019. 3

    2019

  76. [84]

    Robust visual knowledge transfer via extreme learning machine-based domain adaptation

    Lei Zhang and David Zhang. Robust visual knowledge transfer via extreme learning machine-based domain adaptation. IEEE Transactions on Image Processing, 25(10):4959–4973, 2016. 3

    2016

  77. [85]

    Multi-source domain adaptation for semantic segmentation

    Sicheng Zhao, Bo Li, Xiangyu Yue, Yang Gu, Pengfei Xu, Runbo Hu, Hua Chai, and Kurt Keutzer. Multi-source domain adaptation for semantic segmentation. In Advances in Neural Information Processing Systems , pages 7285–7298, 2019. 4

    2019

  78. [86]

    Deep subdomain adaptation network for image classification

    Yongchun Zhu, Fuzhen Zhuang, Jindong Wang, Guolin Ke, Jingwu Chen, Jiang Bian, Hui Xiong, and Qing He. Deep subdomain adaptation network for image classification. IEEE transactions on neural networks and learning systems, 32(4):1713–1722, 2020. 4, 6 Lingkun Luo served as a res...

    2020

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.