Pith. sign in

REVIEW 4 major objections 4 minor 85 references

Decision Boundary Optimization-Informed Domain Adaptation

T0 review · 4 major / 4 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 The paper argues that adding decision-boundary awareness to MMD-based domain adaptation can improve accuracy by up to 9.5 points on standard benchmarks.

desk verdict A plausible incremental MMD-DA extension whose published separation-graph equations have a sign error that flips the intended mechanism; the empirical gains are real but uneven and the theory overclaims. read the letter →

arxiv 2502.06498 v1 pith:YBTPOSQ6 submitted 2025-02-10 cs.CV

Lingkun Luo , Shiqiang Hu , Jie Yang , Liming Chen This is my paper · ORCID
classification cs.CV
keywords domainadaptationmaximummeandiscrepancydecisionboundaryoptimizationcompactinggraphseparationunsuperviseddistributionalignmenttransferlearning
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 proposes a strengthened version of the Maximum Mean Discrepancy (MMD) distance, called Decision Boundary optimization-informed MMD (DB-MMD), for unsupervised domain adaptation. The central idea is to make MMD not only align the source and target distributions but also optimize the classifier's decision boundary by compacting intra-class cross-domain samples and separating inter-class cross-domain samples. The authors claim this allows simultaneous optimization of all three terms of the target error bound, and they demonstrate that embedding DB-MMD into several MMD-based baselines improves accuracy on eight standard datasets, with gains up to 9.5 points.

What carries the argument

The central object is the DB-MMD matrix, which replaces the plain MMD matrix $M_0 + \sum_c M_c$ in the projection objective. It adds two graph-derived matrices: a compacting graph $G_{CG}$ that up-weights pairs of same-class cross-domain samples that are far apart, and a separation graph $G_{SG}$ that up-weights pairs of differently labeled cross-domain samples that are close. The combined matrix is $M_0 + (G_{CG} * \sum_c M_c) - G_{SG} * (M_{S\to T} + M_{T\to S})$, and it is embedded into a generalized eigendecomposition problem to find a shared subspace $A$ that minimizes intra-class distance while maximizing inter-class separation.

What would settle it

On a standard DA benchmark, compute the kernel mean embedding distance between each source subdomain and its pseudo-labeled target subdomain before and after learning the shared subspace; if the distances remain clearly nonzero after optimization, the substitution step in the proof is not satisfied and the claim that DB-MMD provably minimizes all three error-bound terms collapses.

Watch

Extended reading notes

Core claim

The paper claims that a decision-boundary-aware MMD measurement, DB-MMD, can replace plain MMD inside existing domain adaptation methods to achieve better classification on the target domain. DB-MMD is constructed by multiplying the conditional MMD matrix with a compacting graph (to pull together same-class samples across domains, especially those far apart) and adding a separation graph to push apart differently labeled cross-domain samples that lie close to the decision boundary. The paper further claims that this construction enables a single optimization to reduce all terms of the hypothesis error bound $e_T(h) \le e_S(h) + d_H(\mathcal{D}_S, \mathcal{D}_T) + \min\{...\}$, so that the resulting DA model is "theoretically bound guided" and outperforms its MMD-based baselines on eight standard datasets.

Load-bearing premise

The theoretical derivation assumes that after minimizing conditional MMD, every source subdomain and its pseudo-labeled target subdomain have identical kernel mean embeddings, an equality that finite-sample MMD minimization does not actually guarantee.

Editorial extensions

If this is right

  • If DB-MMD works as claimed, existing MMD-based domain adaptation methods can be upgraded just by swapping in the strengthened matrix, without changing their optimization machinery.
  • The improvement of up to 9.5 accuracy points on the PIE dataset suggests that boundary-aware alignment can translate into large gains on tasks with many near-identical classes.
  • Because the compacting and separation graphs are domain-agnostic, the same construction could be applied to other divergence measures (e.g., CORAL) to make them decision-boundary-aware as well.
  • The paper's analysis of convergence and parameter sensitivity suggests that the added graphs do not destabilize training and can even help the model converge in fewer iterations.

Reading between the lines

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

  • A testable extension of the paper's logic: if pseudo-labels on the target domain are noisy, the compacting and separation graphs may amplify errors, so a robust variant would need to down-weight uncertain pseudo-label assignments.
  • The claimed simultaneous optimization of all three error-bound terms is stronger than the experiments alone can establish; the theoretical argument hinges on an equality that finite-sample MMD minimization only approximately satisfies.
  • The paper's comparative results on Office-Home show that shallow DB-MMD models can partly close the gap with deep adversarial DA methods, suggesting that careful distribution alignment can sometimes rival learned feature extraction in DA.
  • The separation graph's reliance on cross-domain inter-class pairs could also be interpreted as a form of hard negative mining, a connection the paper does not explicitly draw.
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

4 major / 4 minor

Summary. The paper proposes DB-MMD, a modified Maximum Mean Discrepancy term that combines marginal/conditional alignment, a cross-subdomain repulsive force, and two learned graphs (a compacting graph and a separation graph) intended to make MMD-based domain adaptation decision-boundary aware. The authors embed DB-MMD into JDA, CDDA, DGA-DA, and MEDA, and report accuracy gains on several benchmarks, including a 9.5-point improvement on CMU PIE and strong Office-Home results. The paper also contains sensitivity analyses, convergence studies, and t-SNE visualizations. The central claims are that DB-MMD simultaneously optimizes all three terms of the Ben-David error bound and that it consistently improves plain-MMD baselines across eight datasets.

Significance. If the proposed mechanism worked as described, a lightweight plug-in that makes MMD alignment boundary-aware would be a useful contribution to shallow unsupervised domain adaptation. The paper has genuine strengths: it evaluates many derived variants across multiple standard benchmarks, includes ablation-style comparisons with and without the compacting/separation graphs, and provides sensitivity and convergence analysis. The central idea of weighting MMD sample pairs by affinity is intuitive and potentially practical. However, the theoretical derivation in Section 3.2.2 relies on an unjustified finite-sample equality, and the matrix implementations of the compacting and separation graphs in Eqs. (15) and (17) have sign inconsistencies that make the published objective behave opposite to the stated intention. In addition, results for four of the eight claimed datasets are deferred to an absent supplement. These issues are load-bearing for the paper's main empirical and theoretical claims, so the contribution is not yet established as stated.

major comments (4)
  1. [3.2.3, Eq. (17)] The separation graph as written implements attraction rather than repulsion of cross-class samples. In Eq. (16), the objective is to minimize W_ij x_i^T x_j over cross-class cross-domain pairs, so close pairs (large W) should receive a large positive penalty when their dot product is positive. In the matrix implementation, the masked entries of B = MS->T + MT->S have value -1/(n_s^c n_t^r), and Eq. (17) sets GSG = -(1/W).*MASK. Hence (GSG.*B)_ij = +1/(W_ij n_s^c n_t^r), which is positive. Substituting into Eq. (19), the maximization term tr(A^T X (GSG.*B) X^T A) is maximized by making x_i^T x_j large and positive for exactly the cross-class pairs that should be separated; equivalently, in the minimization form of Eq. (20) the contribution -GSG.*B is negative, again encouraging positive dot products. The correct sign should be GSG = W.*MASK (up to a class-size normalization), which would make GSG.*B negative and penalize close cross-class pairs most strongly. As published, Eqs. (17)-(20) do not implement the boundary-aware separation mechanism described in the text, so the reported gains cannot be attributed to the stated DB-MMD objective without a corrected equation or released code.
  2. [3.2.3, Eq. (15)] The compacting graph has the same sign problem. Eq. (14) is written as a maximization of (x_i^T x_j)/W_ij, equivalently a minimization of -(x_i^T x_j)/W_ij, so the coefficient of x_i^T x_j inside the final minimization should be negative for intra-class cross-domain pairs. However, Eq. (15) defines GCG = -(1/W).*MASK on entries where Mc has value -1/(n_s^c n_t^c), giving (GCG.*Mc)_ij = +1/(W_ij n_s^c n_t^c). Because this positive coefficient appears inside the minimization in Eq. (19), the optimization encourages negative dot products for same-class cross-domain pairs, pulling intra-class samples apart instead of compacting them. The correct implementation should use GCG = +(1/W).*MASK (up to normalization), so that GCG.*Mc is negative and the minimization shrinks intra-class distances.
  3. [3.2.2, Eq. (12)] The proof that the bidirectional repulsive force of Eq. (11) is equivalent to the single-domain repulsive force of Eq. (10) assumes that minimizing the conditional MMD in Eq. (6.2) yields exact equality of kernel mean embeddings, as stated in Eq. (12). With finite samples, a projection constraint, and regularization, minimizing an empirical MMD reduces the distance but does not force the embeddings to be equal. Therefore the substitution of the target subdomain embedding by the source subdomain embedding in Eq. (11) is not justified, and the claim that Eq. (11) minimizes the same quantity as Eq. (10) is not established. This is load-bearing because Section 3.2.4's assertion that DB-MMD simultaneously optimizes all three terms of the error bound in Eq. (1) relies on this equivalence.
  4. [Section 4.4.3-4.4.6 and Section 4.4.2] The empirical claim of consistent improvement across eight datasets is not fully verifiable from the manuscript. Results for Office+Caltech, COIL20, USPS+MNIST, and VisDA are explicitly deferred to supplemental materials, but no supplement is included. In addition, Section 4.4.2 is titled 'Office-Home' and the text states that ResNet-50 features are used, while the caption of Fig. 7 reads 'Caltech256+Office DATABASE (DeCAF6)'. The reader therefore cannot determine the benchmark/feature configuration that produced the reported Office-Home numbers. These issues should be resolved before the broad empirical claims can be assessed.
minor comments (4)
  1. [Title page and headers] The manuscript still contains template placeholders such as 'SAMPLE JMLR PAPER', 'Author One and Author Two', and 'Editor: My editor'; these should be removed and the author list should be consistent between the title page and the abstract.
  2. [Section 4.5.2] The text says 'λ ∈ {0.1 ≤ k ≤ 10}' when describing the range of the regularization parameter; this should be a range for λ, not k.
  3. [Section 4.3.3 and elsewhere] There are several typos and inconsistent names, e.g., 'Office+HOME' instead of 'Office-Home', 'nowrrow' instead of 'narrow', and 'in par' instead of 'in par with'; a careful proofreading pass is needed.
  4. [Eq. (7) and Eq. (24)] The notation in Eq. (7) is difficult to parse because the subscripts and summation ranges are not cleanly typeset, and in Eq. (24) the matrix M used in the MEDA objective is not explicitly defined in the surrounding text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical gains are benchmark-grounded; proof gaps and sign inconsistencies are correctness issues, not circular reductions.

full rationale

The paper's central claim is empirical: DB-MMD improves MMD-based DA baselines on eight standard benchmarks, and that claim is evaluated against independent test accuracies rather than derived from the method's own fitted values. The theoretical framing in Section 3.2.2 does contain a genuine proof gap: the claimed equivalence of Eq. (11) to Eq. (10) is stated conditional on Eq. (12), the exact equality of kernel mean embeddings after conditional-MMD minimization, which finite-sample MMD minimization does not guarantee. This is a rigor or correctness concern, not a circular reduction, because Eq. (12) is an independent (if unproven) premise rather than the conclusion being established. Likewise, the separation-graph equations appear internally inconsistent: Eq. (16) weights close cross-class pairs by the affinity W, while Eq. (17) sets GSG = -(1/W).*MASK, which would make the implemented term reward large positive dot products on cross-class pairs instead of repelling them. This is a sign/implementation inconsistency and a serious reproducibility risk, but it is not a fit-renamed-as-prediction or a definitional circularity. Self-citations to the authors' earlier CDDA and DGA-DA baselines are used as experimental platforms, but the baselines are published methods with independent benchmark results, and the DB-MMD construction itself is not justified solely by those citations. No load-bearing circular step is exhibited, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central method is a heuristic construction that depends on several fitted hyperparameters and on unproven bridge assumptions: pseudo-label quality, exact equality of kernel means after alignment, and the equivalence of graph weighting to boundary optimization. The benchmark gains are the only independent evidence; there is no parameter-free derivation.

free parameters (5)
  • subspace dimension k = k=100 (k=20 for MEDA+CG on Office+Caltech)
    Chosen by empirical search; sensitivity analysis in Fig.8 shows stability for k at least 100 on PIE tasks.
  • regularization lambda = lambda=0.1 (USPS, MNIST, COIL20, PIE); lambda=1 (Office+Caltech, Office-Home)
    Set per dataset family in Section 4.3.3 without a validation procedure, i.e., by empirical search over target accuracy.
  • manifold trade-off mu = mu=0.01
    Hyperparameter for DGA-DA variants, Section 4.3.3.
  • MEDA hyperparameters alpha, rho, eta = alpha=10, rho=0.1, eta=1
    Set in Section 4.3.3 for MEDA+CG.
  • Gaussian kernel width sigma = not reported
    The affinity matrix W in Eq (4) and the graphs G_CG and G_SG depend on sigma; no value or search range is given, so the graphs are underspecified.
assumptions (5)
  • standard math Ben-David et al. (2010) error bound (Eq. 1) applies to the UDA setting.
    Used to frame DB-MMD as minimizing all three terms; no extension or proof is given.
  • standard math MMD in a universal RKHS measures distribution divergence and its finite-sample estimate is reliable for these datasets.
    Basis of Eqs (2) to (7); standard in the MMD-DA literature.
  • domain assumption NN-assigned pseudo-labels on target sub-domains are accurate enough for conditional MMD and graph masks.
    Section 3.1: target sub-domains require pseudo-labels; noisy labels would misalign the conditional terms.
  • ad hoc to paper Minimizing conditional MMD implies equality of kernel mean embeddings across matched sub-domains (Eq. 12).
    This assumed equality underlies the proof that Eq (11) is equivalent to Eq (10); it is not guaranteed in finite-sample optimization.
  • ad hoc to paper The compacting and separation graphs directly reduce Term 3 (labeling-function discrepancy) of the error bound.
    No formal link is derived; the claim is asserted in Sections 3.2.3 and 3.2.4 and used to support the theoretical-bound narrative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decision Boundary Optimization-Informed Domain Adaptation." pith.science (2026). https://pith.science/paper/YBTPOSQ6

@misc{pith2026250206498,
  author       = {Pith},
  title        = {Pith review of: Decision Boundary Optimization-Informed Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YBTPOSQ6}},
  note         = {Machine review of arXiv:2502.06498}
}
read the original abstract

Maximum Mean Discrepancy (MMD) is widely used in a number of domain adaptation (DA) methods and shows its effectiveness in aligning data distributions across domains. However, in previous DA research, MMD-based DA methods focus mostly on distribution alignment, and ignore to optimize the decision boundary for classification-aware DA, thereby falling short in reducing the DA upper error bound. In this paper, we propose a strengthened MMD measurement, namely, Decision Boundary optimization-informed MMD (DB-MMD), which enables MMD to carefully take into account the decision boundaries, thereby simultaneously optimizing the distribution alignment and cross-domain classifier within a hybrid framework, and leading to a theoretical bound guided DA. We further seamlessly embed the proposed DB-MMD measurement into several popular DA methods, e.g., MEDA, DGA-DA, to demonstrate its effectiveness w.r.t different experimental settings. We carry out comprehensive experiments using 8 standard DA datasets. The experimental results show that the DB-MMD enforced DA methods improve their baseline models using plain vanilla MMD, with a margin that can be as high as 9.5.

Figures

Figures reproduced from arXiv: 2502.06498 by the authors.

Figure 1
Figure 1. Fig.1.(a) shows that the source domain and the target domain samples depict a large domain divergence in the original feature space. Fig.1.(b) highlights that distribution alignment-based DA drags close the domains and the sub-domains but tends to ignore to optimize the decision boundary for yielding the classifier optimization ensured func￾tional learning as required in Fig.1.(c). reduction than the decision bounda… view at source ↗
Figure 2
Figure 2. Illustration of the proposed decision boundary optimization-informed DA (DB-DA). Fig.2 (a): the original source and target domain distributions; Fig.2 (b,c) illustrate DB-DA aligning cross-domain distributions closely yet discriminatively by using MMD. Fig.2 (d) shows the proposed DA aware of decision boundary through the specifically designed ’compacting graph’ and ’separation graph’; Fig.2 (e) illustrates the achi… view at source ↗
Figure 3
Figure 3. In Fig.3.(a), DA explores the effectiveness of distribution alignment to drag close the domains and the sub-domains, while ignoring to optimize the samples lying around deci￾sion boundaries (Fig.3.(b,c)) for generating a decision boundary optimization guaran￾teed functional learning as illustrated in Fig.3.(d). To fight the aforementioned weaknesses, we introduce here the decision boundary aware graph. It consists o… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Sample images from 8 datasets used in our experiments. Each dataset represents a dif￾ferent domain. The OFFICE dataset contains three sub-datasets: DSLR, Amazon, and Webcam. Jhuo et al. (2012); (21) LPJT Li et al. (2019); (22) DGA-DA Luo et al. (2020); (23) GEF through…
Figure 5
Figure 5. Figure 5: In Fig.5.(a), the red portion denotes the baseline model of JDA, which is further improved by CDDA in hybridizing the repulse force term formalized in the yellow part. Then, based on CDDA, a geometric regularization is also embedded to formalize the DGA￾DA. Fig.5.(b) i…
Figure 6
Figure 6. Figure 6: Accuracy% on the PIE Images Dataset. 4.4.2 EXPERIMENTS ON THE OFFICE-HOME DATASET As introduced in DAHVenkateswara et al. (2017), Office-Home is a novel challenging benchmark for the DA task. It contains 4 different domains with 65 object categories, thereby generating…
Figure 7
Figure 7. Figure 7: Accuracy% on the Office-Home Images Dataset. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Sensitivity analysis of the proposed methods, i.e., DGA-DA, DGA-DA+CG and DGA￾DA+DB, using PIE dataset w.r.t. subspace dimension k. Interestingly, in Fig.8.(c) and Fig.8.(f), the performance of DGA-DA doesn’t achieve stability until k reaches 160, while the derived mod…
Figure 9
Figure 9. Figure 9: The classification accuracies of the proposed DGA-DA, DGA-DA+CG and DGA￾DA+DB methods vs. the parameter λ on the selected three cross domains data sets. In Fig.9, we plot the classification accuracy of the proposed DA methods w.r.t different values of λ on the Office+C…
Figure 10
Figure 10. Figure 10: Detailed discussion of the proposed decision boundary aware mechanism w.r.t different manifold learning strategies. Manifold learning techniques have been widely applied in DA algorithms for lifting the per￾formance in solving cross-domain tasks, while preserving the …
Figure 11
Figure 11. Figure 11: Convergence analysis using 6 cross-domain image classification tasks on the Of￾fice+Caltech dataset. (accuracy w.r.t #iterations) 4.5.5 T-SNE VISUALIZATION Using the PIE dataset and t-SNE visualization Van der Maaten and Hinton (2008) method, Fig.12 visualizes the cla…
Figure 12
Figure 12. Figure 12: Accuracy(%) and Visualization results of the PIE-9 → PIE-27 DA task. Fig.12(a), Fig.12(b), and Fig.12(c) are visualization results of PIE-9, PIE-27, and PIE-27&9 datasets in their Original data space, respectively. Fig.12(d) visualizes both the source and target datas…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

85 extracted references · 64 canonical work pages

  1. [1]

    Laplacian eigenmaps for dimensionality reduction and data representation

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation. Neural computation, 15 0 (6): 0 1373--1396, 2003

    2003

  2. [2]

    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 0 (1): 0 151--175, 2010

    2010

  3. [3]

    Integrating structured biological data by kernel maximum mean discrepancy

    Karsten M Borgwardt, Arthur Gretton, Malte J Rasch, Hans-Peter Kriegel, Bernhard Sch \"o lkopf, and Alex J Smola. Integrating structured biological data by kernel maximum mean discrepancy. Bioinformatics, 22 0 (14): 0 e49--e57, 2006

    2006

  4. [4]

    Domain separation networks

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

    2016

  5. [5]

    Learning smooth representation for unsupervised domain adaptation

    Guanyu Cai, Lianghua He, MengChu Zhou, Hesham Alhumade, and Die Hu. Learning smooth representation for unsupervised domain adaptation. IEEE Transactions on Neural Networks and Learning Systems, 2021

    2021

  6. [6]

    Weinberger, and Fei Sha

    Minmin Chen, Zhixiang Eddie Xu, Kilian Q. Weinberger, and Fei Sha. Marginalized denoising autoencoders for domain adaptation. CoRR, abs/1206.4683, 2012. URL http://arxiv.org/abs/1206.4683

    arXiv 2012

  7. [7]

    A graph embedding framework for maximum mean discrepancy-based domain adaptation algorithms

    Yiming Chen, Shiji Song, Shuang Li, and Cheng Wu. A graph embedding framework for maximum mean discrepancy-based domain adaptation algorithms. IEEE Trans. Image Process. , 29: 0 199--213, 2020. doi:10.1109/TIP.2019.2928630. URL https://doi.org/10.1109/TIP.2019.2928630

    work page arXiv 2020

  8. [8]

    Mcdal: Maximum classifier discrepancy for active learning

    Jae Won Cho, Dong-Jin Kim, Yunjae Jung, and In So Kweon. Mcdal: Maximum classifier discrepancy for active learning. IEEE Transactions on Neural Networks and Learning Systems, 2022

    work page 2022

Show all 85 references
  1. [9]

    Joint distribution optimal transportation for domain adaptation

    Nicolas Courty, R \'e mi Flamary, Amaury Habrard, and Alain Rakotomamonjy. Joint distribution optimal transportation for domain adaptation. In Advances in Neural Information Processing Systems, pages 3733--3742, 2017 a

    2017

  2. [10]

    Optimal transport for domain adaptation

    Nicolas Courty, R \'e mi Flamary, Devis Tuia, and Alain Rakotomamonjy. Optimal transport for domain adaptation. IEEE transactions on pattern analysis and machine intelligence, 39 0 (9): 0 1853--1865, 2017 b

    2017

  3. [11]

    Robust transfer metric learning for image classification

    Zhengming Ding and Yun Fu. Robust transfer metric learning for image classification. IEEE Trans. Image Processing , 26 0 (2): 0 660--670, 2017. doi:10.1109/TIP.2016.2631887. URL https://doi.org/10.1109/TIP.2016.2631887

    2017

  4. [12]

    Unsupervised visual domain adaptation using subspace alignment

    Basura Fernando, Amaury Habrard, Marc Sebban, and Tinne Tuytelaars. Unsupervised visual domain adaptation using subspace alignment. In Proceedings of the IEEE international conference on computer vision, pages 2960--2967, 2013

    2013

  5. [13]

    Augmented Lagrangian methods: applications to the numerical solution of boundary-value problems, volume 15

    Michel Fortin and Roland Glowinski. Augmented Lagrangian methods: applications to the numerical solution of boundary-value problems, volume 15. Elsevier, 2000

    2000

  6. [14]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Fran c ois Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial training of neural networks. The Journal of Machine Learning Research, 17 0 (1): 0 2096--2030, 2016

    2016

  7. [15]

    Bastiaan Kleijn, and Mengjie Zhang

    Muhammad Ghifary, David Balduzzi, W. Bastiaan Kleijn, and Mengjie Zhang. Scatter component analysis: A unified framework for domain adaptation and domain generalization. IEEE Trans. Pattern Anal. Mach. Intell. , 39 0 (7): 0 1414--1430, 2017. doi:10.1109/TPAMI.2016.2599532. URL...

    2017

  8. [16]

    Geodesic flow kernel for unsupervised domain adaptation

    Boqing Gong, Yuan Shi, Fei Sha, and Kristen Grauman. Geodesic flow kernel for unsupervised domain adaptation. In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on, pages 2066--2073. IEEE, 2012

    2012

  9. [17]

    Connecting the dots with landmarks: Discriminatively learning domain-invariant features for unsupervised domain adaptation

    Boqing Gong, Kristen Grauman, and Fei Sha. Connecting the dots with landmarks: Discriminatively learning domain-invariant features for unsupervised domain adaptation. In International conference on machine learning, pages 222--230. PMLR, 2013

    2013

  10. [18]

    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

    2014

  11. [19]

    Deep learning

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

    2016

  12. [20]

    A kernel method for the two-sample-problem

    Arthur Gretton, Karsten Borgwardt, Malte Rasch, Bernhard Sch \"o lkopf, and Alex Smola. A kernel method for the two-sample-problem. Advances in neural information processing systems, 19, 2006

    2006

  13. [21]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

    2016

  14. [22]

    Knowledge distillation with adversarial samples supporting decision boundary

    Byeongho Heo, Minsik Lee, Sangdoo Yun, and Jin Young Choi. Knowledge distillation with adversarial samples supporting decision boundary. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pages 3771--3778, 2019

    2019

  15. [23]

    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

    2017

  16. [24]

    C y CADA : Cycle-consistent adversarial domain adaptation

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

    1989

  17. [25]

    Unsupervised domain adaptation with hierarchical gradient synchronization

    Lanqing Hu, Meina Kan, Shiguang Shan, and Xilin Chen. Unsupervised domain adaptation with hierarchical gradient synchronization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4043--4052, 2020

    2020

  18. [26]

    Jonathan J. Hull. A database for handwritten text recognition research. IEEE Trans. Pattern Anal. Mach. Intell. , 16 0 (5): 0 550--554, 1994. doi:10.1109/34.291440. URL https://doi.org/10.1109/34.291440

    1994 doi

  19. [27]

    Robust visual domain adaptation with low-rank reconstruction

    I-Hong Jhuo, Dong Liu, DT Lee, and Shih-Fu Chang. Robust visual domain adaptation with low-rank reconstruction. In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on, pages 2168--2175. IEEE, 2012

    2012

  20. [28]

    Optimal bayesian transfer learning

    Alireza Karbalayghareh, Xiaoning Qian, and Edward R Dougherty. Optimal bayesian transfer learning. IEEE Transactions on Signal Processing, 66 0 (14): 0 3724--3739, 2018

    2018

  21. [29]

    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, 2004

    2004

  22. [30]

    Unsupervised visual domain adaptation: A deep max-margin gaussian process approach

    Minyoung Kim, Pritish Sahu, Behnam Gholami, and Vladimir Pavlovic. Unsupervised visual domain adaptation: A deep max-margin gaussian process approach. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4380--4390, 2019

    2019

  23. [31]

    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

    2012

  24. [32]

    Iterative label cleaning for transductive and semi-supervised few-shot learning

    Michalis Lazarou, Tania Stathaki, and Yannis Avrithis. Iterative label cleaning for transductive and semi-supervised few-shot learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8751--8760, 2021

    2021

  25. [33]

    Gradient-based learning applied to document recognition

    Yann LeCun, L \'e on Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86 0 (11): 0 2278--2324, 1998

    1998

  26. [34]

    Dranet: Disentangling representation and adaptation networks for unsupervised cross-domain adaptation

    Seunghun Lee, Sunghyun Cho, and Sunghoon Im. Dranet: Disentangling representation and adaptation networks for unsupervised cross-domain adaptation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15252--15261, 2021

    2021

  27. [35]

    Cross-domain adaptive clustering for semi-supervised domain adaptation

    Jichang Li, Guanbin Li, Yemin Shi, and Yizhou Yu. Cross-domain adaptive clustering for semi-supervised domain adaptation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2505--2514, 2021 a

    2021

  28. [36]

    Locality preserving joint transfer for domain adaptation

    Jingjing Li, Mengmeng Jing, Ke Lu, Lei Zhu, and Heng Tao Shen. Locality preserving joint transfer for domain adaptation. IEEE Transactions on Image Processing, 28 0 (12): 0 6103--6115, 2019

    2019

  29. [37]

    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 0 (11): 0 3918--3930, 2020 a

    2020

  30. [38]

    Enhanced transport distance for unsupervised domain adaptation

    Mengxue Li, Yi-Ming Zhai, You-Wei Luo, Peng-Fei Ge, and Chuan-Xian Ren. Enhanced transport distance for unsupervised domain adaptation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13936--13944, 2020 b

    2020

  31. [39]

    Bi-classifier determinacy maximization for unsupervised domain adaptation

    Shuang Li, Fangrui Lv, Binhui Xie, Chi Harold Liu, Jian Liang, and Chen Qin. Bi-classifier determinacy maximization for unsupervised domain adaptation. In AAAI, 2021 b

    2021

  32. [40]

    Domain adaptation with auxiliary target domain-oriented classifier

    Jian Liang, Dapeng Hu, and Jiashi Feng. Domain adaptation with auxiliary target domain-oriented classifier. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16632--16642, 2021

    2021

  33. [41]

    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

    2013

  34. [42]

    Mingsheng Long, Jianmin Wang, Guiguang Ding, Sinno Jialin Pan, and Philip S. Yu. Adaptation regularization: A general framework for transfer learning. IEEE Trans. Knowl. Data Eng. , 26 0 (5): 0 1076--1089, 2014 a . doi:10.1109/TKDE.2013.111. URL https://doi.org/10.1109/TKDE.2013.111

    2014 doi

  35. [43]

    Transfer joint matching for unsupervised domain adaptation

    Mingsheng Long, Jianmin Wang, Guiguang Ding, Jiaguang Sun, and Philip S Yu. Transfer joint matching for unsupervised domain adaptation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1410--1417, 2014 b

    2014

  36. [44]

    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

    2015

  37. [45]

    Deep transfer learning with joint adaptation networks

    Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I Jordan. Deep transfer learning with joint adaptation networks. In International conference on machine learning, pages 2208--2217. PMLR, 2017

    2017

  38. [46]

    Mingsheng Long, Zhangjie Cao, Jianmin Wang, and Michael I. Jordan. Conditional adversarial domain adaptation. In Neural Information Processing Systems 2018, NeurIPS 2018,, pages 1647--1657, 2018. URL https://proceedings.neurips.cc/paper/2018/hash/ab88b15733f543179858600245108d...

    2018

  39. [47]

    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

    2018

  40. [48]

    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 0 (2): 0 37:1--37:35, 2020. doi:10.1145/3379344. URL https://doi.org/10.1145/3379344

    2020 doi

  41. [49]

    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. URL http://arxiv.org/abs/1705.08620

    2017 arXiv

  42. [50]

    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

    2020

  43. [51]

    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: 0 7322--7337, 2022

    2022

  44. [52]

    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, pages 1--24, 2023

    2023

  45. [53]

    Ng, Michael I

    Andrew Y. Ng, Michael I. Jordan, and Yair Weiss. On spectral clustering: Analysis and an algorithm. In T. G. Dietterich, S. Becker, and Z. Ghahramani, editors, Advances in Neural Information Processing Systems 14, pages 849--856. MIT Press, 2002. URL http://papers.nips.cc/pape...

    2002

  46. [54]

    A survey on transfer learning

    Sinno Jialin Pan and Qiang Yang. A survey on transfer learning. IEEE Transactions on knowledge and data engineering, 22 0 (10): 0 1345--1359, 2010

    2010

  47. [55]

    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 0 (2): 0 199--210, 2011

    2011

  48. [56]

    V. M. Patel, R. Gopalan, R. Li, and R. Chellappa. Visual domain adaptation: A survey of recent advances. IEEE Signal Processing Magazine, 32 0 (3): 0 53--69, May 2015. ISSN 1053-5888. doi:10.1109/MSP.2014.2347059

    2015

  49. [57]

    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

    2018

  50. [58]

    Visda: The visual domain adaptation challenge

    Xingchao Peng, Ben Usman, Neela Kaushik, Judy Hoffman, Dequan Wang, and Kate Saenko. Visda: The visual domain adaptation challenge. arXiv preprint arXiv:1710.06924, 2017

    2017 arXiv

  51. [59]

    Max-margin markov networks

    BTCGD Roller, C Taskar, and D Guestrin. Max-margin markov networks. Advances in neural information processing systems, 16: 0 25, 2004

    2004

  52. [60]

    Beyond sharing weights for deep domain adaptation

    Artem Rozantsev, Mathieu Salzmann, and Pascal Fua. Beyond sharing weights for deep domain adaptation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018

    2018

  53. [61]

    Asymmetric tri-training for unsupervised domain adaptation

    Kuniaki Saito, Yoshitaka Ushiku, and Tatsuya Harada. Asymmetric tri-training for unsupervised domain adaptation. In International conference on machine learning, pages 2988--2997. PMLR, 2017

    2017

  54. [62]

    Maximum classifier discrepancy for unsupervised domain adaptation

    Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku, and Tatsuya Harada. Maximum classifier discrepancy for unsupervised domain adaptation. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018 , pages 3723--3732. ...

    2018

  55. [63]

    o lkopf, Alexander J. Smola, and Klaus - Robert M \

    Bernhard Sch \" o lkopf, Alexander J. Smola, and Klaus - Robert M \" u ller. Nonlinear component analysis as a kernel eigenvalue problem. Neural Computation, 10 0 (5): 0 1299--1319, 1998. doi:10.1162/089976698300017467. URL https://doi.org/10.1162/089976698300017467

    1998 doi

  56. [64]

    Learning transferrable representations for unsupervised domain adaptation

    Ozan Sener, Hyun Oh Song, Ashutosh Saxena, and Silvio Savarese. Learning transferrable representations for unsupervised domain adaptation. In Advances in Neural Information Processing Systems, pages 2110--2118, 2016

    2016

  57. [65]

    Transfer learning for visual categorization: A survey

    Ling Shao, Fan Zhu, and Xuelong Li. Transfer learning for visual categorization: A survey. IEEE Trans. Neural Netw. Learning Syst. , 26 0 (5): 0 1019--1034, 2015. doi:10.1109/TNNLS.2014.2330900. URL https://doi.org/10.1109/TNNLS.2014.2330900

    2015

  58. [66]

    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 0 (1-2): 0 74--93, 2014. doi:10.1007/s11263-014-0696-6. URL http://dx.doi.org/10.1007/s11263-014-0696-6

    2014 doi

  59. [67]

    S. Si, D. Tao, and B. Geng. Bregman divergence-based regularization for transfer subspace learning. IEEE Transactions on Knowledge and Data Engineering, 22 0 (7): 0 929--942, July 2010. ISSN 1041-4347. doi:10.1109/TKDE.2009.126

    2010 doi

  60. [68]

    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

    2016

  61. [69]

    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

    2016

  62. [70]

    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. URL http://arxiv.org/abs/1412.3474

    2014 arXiv

  63. [71]

    Adversarial discriminative domain adaptation

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

    2017

  64. [72]

    Muhammad Uzair and Ajmal S. Mian. Blind domain adaptation with augmented extreme learning machine features. IEEE Trans. Cybernetics , 47 0 (3): 0 651--660, 2017. doi:10.1109/TCYB.2016.2523538. URL https://doi.org/10.1109/TCYB.2016.2523538

    2017

  65. [73]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9 0 (11), 2008

    2008

  66. [74]

    Deep hashing network for unsupervised domain adaptation

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep hashing network for unsupervised domain adaptation. arXiv preprint arXiv:1706.07522, 2017

    2017 arXiv

  67. [75]

    Cross-domain metric learning based on information theory

    Hao Wang, Wei Wang, Chen Zhang, and Fanjiang Xu. Cross-domain metric learning based on information theory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 28, 2014

    2014

  68. [76]

    Visual domain adaptation with manifold embedded distribution alignment

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

    2018

  69. [77]

    Transfer learning with dynamic distribution adaptation

    Jindong Wang, Yiqiang Chen, Wenjie Feng, Han Yu, Meiyu Huang, and Qiang Yang. Transfer learning with dynamic distribution adaptation. ACM Transactions on Intelligent Systems and Technology (TIST), 11 0 (1): 0 1--25, 2020

    2020

  70. [78]

    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

    2019

  71. [79]

    Toalign: Task-oriented alignment for unsupervised domain adaptation

    Guoqiang Wei, Cuiling Lan, Wenjun Zeng, and Zhibo Chen. Toalign: Task-oriented alignment for unsupervised domain adaptation. arXiv preprint arXiv:2106.10812, 2021

    2021 arXiv

  72. [80]

    Entropy minimization versus diversity maximization for domain adaptation

    Xiaofu Wu, Suofei Zhang, Quan Zhou, Zhen Yang, Chunming Zhao, and Longin Jan Latecki. Entropy minimization versus diversity maximization for domain adaptation. IEEE Transactions on Neural Networks and Learning Systems, 2021

    2021

  73. [81]

    Discriminative transfer subspace learning via low-rank and sparse representation

    Yong Xu, Xiaozhao Fang, Jian Wu, Xuelong Li, and David Zhang. Discriminative transfer subspace learning via low-rank and sparse representation. IEEE Trans. Image Processing , 25 0 (2): 0 850--863, 2016. doi:10.1109/TIP.2015.2510498. URL https://doi.org/10.1109/TIP.2015.2510498

    2016

  74. [82]

    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

    2020

  75. [83]

    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

    2017

  76. [84]

    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

    2019

  77. [85]

    Unpaired image-to-image translation using cycle-consistent adversarial networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2223--2232, 2017

    2017

Pith tools

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