Pith. sign in

REVIEW 5 major objections 5 minor 66 references

Unveiling the Superior Paradigm: A Comparative Study of Source-Free Domain Adaptation and Unsupervised Domain Adaptation

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Source-free domain adaptation generally outperforms classical unsupervised domain adaptation in real-world settings, on grounds of efficiency, privacy, and robustness against negative transfer.

desk verdict A useful new data-model fusion scenario and a small, sensible weighting method sit inside an overclaimed comparison paper; the headline 'SFDA is superior' is not backed by controlled evidence. read the letter →

arxiv 2411.15844 v1 pith:QSNLQRLV submitted 2024-11-24 cs.LG cs.AIcs.CV

Fan Wang , Zhongyi Han , Xingbo Liu , Xin Gao , Yilong Yin This is my paper · ORCID
classification cs.LGcs.AIcs.CV
keywords source-freedomainadaptationunsupervisednegativetransferpredictivecodingtheorymulti-sourcedata-modelfusionweightestimationshift
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 argues that source-free domain adaptation (SFDA), which adapts a pre-trained model to a new target domain without seeing the original training data, generally outperforms classical unsupervised domain adaptation (UDA), which re-aligns distributions using both source and target data. The authors ground this claim in predictive coding theory—the idea that efficient learning proceeds by updating an existing internal model to correct prediction errors—and support it with benchmark experiments across Office-31, Office-Home, VLCS, TerraIncognita, and DomainNet. They report that SFDA converges in roughly 200 iterations versus 1,000 to 5,000 for UDA, needs far less storage, focuses on the target domain instead of interpolating source and target distributions, and resists negative transfer when source and target domains differ substantially. The paper also introduces a data-model fusion scenario, in which some parties share raw data while others share only pre-trained models, and proposes a Model Estimation and Adaptation (MEA) method that weights source models using their accuracy on visible source domains plus target-domain confidence. If the claim holds, practitioners should favor source-free methods in privacy-sensitive and resource-limited deployments.

What carries the argument

The load-bearing object is the pre-trained source model treated as a fixed prior, together with the weight-estimation rule inside the MEA framework. Predictive coding theory is used as an analogy: the source model is the brain's internal model, target data are sensory inputs, entropy or pseudo-label confidence acts as prediction error, and self-supervised or contrastive losses update the model. In the MEA framework the central identity is the combined weight $w_i = w_t^i + \lambda w_s^i$, where $w_t^i$ is the normalized average softmax confidence of source model $i$ on target samples and $w_s^i$ is its normalized accuracy on visible source domains other than its own. This proxy-based weighting replaces equal or target-only weighting, and the paper claims it better estimates each source model's contribution in the data-model fusion setting.

What would settle it

Run representative UDA and SFDA methods (e.g., DANN, MCD, MDD versus SHOT, NRC, AaD) under identical data splits, backbone, and training budget on Office-Home, DomainNet, and TerraIncognita; if UDA matches or exceeds SFDA on most target tasks, the paper's 'generally outperforms' claim would be undercut. Similarly, if adding a dissimilar source domain's data to an SFDA objective ever improves target accuracy on large-gap tasks, the proposed negative-transfer mechanism would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that SFDA is not merely a privacy-preserving fallback but the superior paradigm for real-world domain adaptation. Because SFDA optimizes only a target-domain objective using the frozen source model as a prior, it avoids learning interpolated distributions that mix dissimilar source and target data, which is why it suffers less negative transfer and overfitting. Empirically, SFDA methods (SHOT, NRC, AaD and their multi-source variants) reach stable target accuracy within about 200 iterations, while UDA methods (DAN, DANN, MCD, MDD, MFSAN, and others) need 1,000 to 5,000 iterations, and SFDA dominates on large-gap benchmarks such as DomainNet and TerraIncognita. The authors also define data-model fusion, a previously unnamed setting where some stakeholders provide labeled source data and others provide only pre-trained models; standard UDA cannot use models and standard SFDA cannot use data in that setting. Their MEA framework adapts multi-SFDA baselines such as SHOTavg and DATE by estimating per-model weights from proxy accuracy on visible source domains plus average target confidence, yielding average gains of 0.5% over SHOTavg and 0.9% over DATE on DomainNet while outperforming multi-UDA methods by several accuracy points.

Load-bearing premise

The conclusion that SFDA generally outperforms UDA rests on the assumption that the reported numbers and learning curves from different methods are comparable even though only some results were reproduced under identical protocols, while others were taken from their original publications.

Editorial extensions

If this is right

  • Practitioners in privacy-constrained or storage-limited settings can expect SFDA to deliver competitive or better accuracy with roughly 5 to 25 times less training time and orders of magnitude less storage.
  • On datasets with large domain gaps, a target-only learning objective should reduce negative transfer, so teams should weigh source-data access against the risk of interpolating incompatible distributions.
  • The MEA weight estimator provides a plug-in upgrade for multi-SFDA pipelines, with consistent per-domain accuracy gains on DomainNet for both SHOTavg and DATE baselines.
  • Standard multi-UDA methods cannot exploit a mix of shared data and shared models, so the data-model fusion setting requires an SFDA-style framework to use all available resources.
  • If SFDA indeed converges faster and resists overfitting, then adaptation budgets in real deployments should be reallocated toward target-side validation rather than source retraining.

Reading between the lines

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

  • The predictive coding connection is presented as an analogy rather than a formal proof; the paper's case stands or falls on the empirical comparisons, not on the theory alone.
  • The proxy-accuracy idea assumes that performance on other visible source domains measures a model's relevance to the target domain; a testable extension is to replace proxy accuracy with a transferability metric computed directly between source model features and target features.
  • A natural extension is to apply MEA to medical imaging or fraud detection, where some institutions release models and others release data; the expected benefit is a few percent accuracy gain plus privacy compliance.
  • The data-model fusion scenario generalizes both UDA and SFDA: when no source data is visible it reduces to multi-SFDA, and when all source data is visible it approaches MUDA with extra source models, so a unified benchmark could emerge from this framing.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper compares Unsupervised Domain Adaptation (UDA) and Source-Free Domain Adaptation (SFDA) and claims that SFDA generally outperforms UDA in real-world scenarios, based on predictive coding theory and experiments on Office-Home, DomainNet, TerraIncognita, and other benchmarks. The authors further introduce a 'data-model fusion' scenario in which some stakeholders provide raw data and others provide only pre-trained models, and propose a Model Estimation and Adaptation (MEA) method that weights source models using visible source data as proxies. The MEA method is evaluated on DomainNet with two multi-SFDA baselines, reporting small average gains over those baselines.

Significance. If the central claim were established, the paper would provide practically useful guidance: practitioners could prefer SFDA in privacy- and resource-constrained settings, and the MEA method would offer a way to exploit mixed data/model availability. The paper also assembles a broad set of UDA and SFDA baselines and highlights an interesting, underexplored scenario. However, the empirical evidence as presented does not yet support the paradigm-level conclusion: many baseline numbers are sourced from different publications with different protocols, no error bars or repeated runs are reported, and no code is available. The predictive-coding discussion is an analogy rather than a formal theory. These limitations are serious enough that the current claims should not be taken as established.

major comments (5)
  1. [Section III-A3, Fig. 2] The headline claim that 'SFDA generally outperforms UDA' rests on cross-paradigm comparisons that are not controlled. Section III-A3 states that only some results are reproduced, and the Fig. 2 caption explicitly says 'we reproduce the DomainNet and Terra results... and record other results as per the original publication.' UDA numbers taken from different papers may use different data splits, backbone initializations, training budgets, and hyperparameters, so the observed accuracy gaps could be implementation artifacts rather than paradigm-level differences. The authors should rerun all compared methods under a single protocol and report mean and standard deviation over multiple seeds.
  2. [Section III-C1, Eqs. (1)-(2), Fig. 1] The time-efficiency advantage is largely definitional and confounded by initialization. Equations (1) and (2) state T(A) ≪ T(RA), meaning SFDA needs only adaptation time while UDA needs retraining with source data. This inequality follows from the problem setup, not from an empirical discovery about the methods: UDA methods are typically trained end-to-end from an ImageNet-initialized backbone using both source and target data, whereas SFDA starts from a fully trained source model and only runs target-side adaptation. Fig. 1 therefore reflects the different starting points and training protocols, not an intrinsic superiority of SFDA. The paper should either control for total compute including source training or explicitly reframe the time advantage as a property of the setting, not as evidence that SFDA methods are better.
  3. [Section III-C2, Section III-C3, Figs. 3-5] The learning-objective and negative-transfer experiments do not provide a valid comparison between UDA and SFDA. The base method is an SFDA model, and the 'Expanded Base' variants add source-data losses to that SFDA model; this is not equivalent to training a UDA method with its standard joint alignment objective from scratch. UDA methods such as DANN and MCD are designed to align distributions through adversarial or discrepancy losses during end-to-end training, so the fact that naively adding a dissimilar source loss to an SFDA pipeline hurts performance does not demonstrate UDA's inferiority. The authors should compare actual UDA methods against actual SFDA methods under matched training budgets and schedules, or explicitly restrict their claims to an ablation of source-data availability within an SFDA pipeline.
  4. [Section III-B] The predictive-coding theory is used as a post-hoc analogy rather than a formal theoretical argument. The section describes a mapping between predictive coding and SFDA but provides no formal model, no assumptions, and no testable quantitative predictions that would distinguish SFDA from UDA. Since the Abstract claims the paper demonstrates SFDA superiority 'through predictive coding theory,' the authors should clarify the precise sense in which the theory predicts the observed empirical ordering, or soften the claim to say that predictive coding provides an interpretative framework. As written, the theoretical analysis does not load-bear for the central claim.
  5. [Section IV-B4, Table I] The evaluation of the proposed MEA framework is preliminary and lacks statistical validation. Table I reports only single accuracy values, with no standard deviations, no significance tests, no sensitivity analysis for the hyperparameter λ in Eq. (9), and no specification of which two source domains were 'randomly selected' as visible sources. The average gains over SHOTavg and DATE are +0.5% and +0.9%, respectively, which may be within run-to-run noise for deep domain adaptation. The authors should provide multi-seed results, error bars, an ablation of the proxy-weight component, and a precise description of the data-model split used in the experiments.
minor comments (5)
  1. [Section III-C1] The statement that a source model requires 'at least 40 orders of magnitude less' storage than source data appears to be a typo, since 40 orders of magnitude is 10^40; if the intended claim is '40 times less' or 'several orders of magnitude less,' it should be corrected.
  2. [Throughout] The paper uses inconsistent method names, including 'SHOT Avg' vs. 'SHOTavg,' 'Date' vs. 'DA TE,' and 'onBias' vs. 'OnBias'; these should be normalized to the names used in the original publications.
  3. [Section III-C4, Fig. 6] The caption states that the training-to-test ratio is 9:1, but the text does not explain how the target data were split into training and test sets; this detail should be specified in the implementation details.
  4. [Table I] The column labeled 'SF' uses the value 'partial' for MEA methods, but the caption says 'SF denotes source data free'; the meaning of 'partial' should be clarified in the table caption or a footnote.
  5. [Section III-A3] The paper states that a code release will follow publication, but no code or reproducibility package is currently available; given the reliance on many external baselines, the authors should consider making at least the reproduction scripts available in a supplementary document.

Circularity Check

3 steps flagged · score 4.0 of 10

Several advertised SFDA advantages (time, storage, learning objective, negative transfer) are definitional consequences of the paradigm choice; the accuracy comparisons remain externally grounded, so circularity is partial, not total.

  1. self definitional [Section III-C1, Eq. (1).]
    "Due to the need for retraining from scratch, UDA generally takes significantly longer than SFDA: T(A) ≪ T(RA)."

    This inequality is entailed by the definitions of the two paradigms given in the paper: SFDA adapts a pre-trained source model without accessing source data, while UDA aligns distributions using source data. A method that by definition omits source-data retraining will necessarily spend less time on source-side training. The paper nevertheless presents time efficiency (and, in the following paragraph, the storage advantage of a model over a dataset) as an empirically supported advantage of SFDA. The conclusion is already contained in the setup, so it cannot independently demonstrate SFDA's superiority.

  2. self definitional [Section III-C2, Eqs. (3)-(4).]
    "Eq. (3) shows that UDA prioritizes source data and treats target data as auxiliary, relying on distribution alignment to identify target samples when source and target domains are similar. In contrast, Eq. (4) highlights SFDA's focus on utilizing the source model's output to capture the target domain's structural characteristics."

    Equations (3) and (4) are the authors' own formal definitions of the UDA and SFDA objectives, not derived results. The claimed 'targeted learning objective' advantage of SFDA is read directly from these definitions: the UDA loss is written with an explicit source-data term and the SFDA loss is written without one. The subsequent experiments add source data to an SFDA baseline ('Expanded Base') rather than training a genuine UDA method from scratch, so the observed performance gap reflects the definitional choice of what was labeled UDA versus SFDA, not an independent empirical discovery.

1 more flagged steps
  1. self definitional [Section III-C3, Negative transfer analysis.]
    "In scenarios with significant source-target distribution disparities, the source model demonstrates greater resilience to negative transfer compared to source data."

    Because SFDA is defined as adapting with no access to source data, it cannot suffer negative transfer from source data by construction; the 'resilience' is a logical consequence of the setting rather than an observed property. The supporting experiment (Fig. 5) takes MSFDA methods and then adds visible source data from a dissimilar domain, which is a constructed modification of the SFDA setting, not a comparison against an actual UDA method. The negative-transfer 'advantage' therefore reduces to the definitional difference between the paradigms.

full rationale

The paper's accuracy comparison between SFDA and UDA is largely external: Table I and Fig. 2 compare reported/reproduced benchmark numbers, and the MEA weight estimation (Eqs. 7-9) uses proxy source-domain accuracy and target confidence as independent signals. That part of the paper is not circular. However, the paper frames SFDA as 'generally outperforming' UDA using several advantages that are definitional consequences of the paradigm definitions: time efficiency (Eq. 1), storage (model versus dataset), a 'targeted' learning objective (Eq. 3 vs. Eq. 4), and reduced negative transfer from source data all follow from the fact that SFDA never touches source data. The learning-objective and negative-transfer experiments also modify an SFDA baseline by adding source losses rather than training real UDA methods, so those demonstrations are constructed from the same definitional gap. The protocol issue that many UDA baselines are recorded 'as per the original publication' rather than reproduced is a correctness and comparability risk, not a circularity, and is not counted in the score. Overall circularity is moderate: some advertised advantages reduce by construction, but the central accuracy claim retains independent external grounding, so a score of 4 is appropriate.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim depends on comparability of external baselines and on two proxy assumptions for MEA. The explicit hyperparameter lambda in Eq. (9) is unreported, and the predictive-coding framing is an analogy rather than a derivation.

free parameters (1)
  • lambda (weight balance in Eq. 9) = not reported
    Balances proxy accuracy weight w^s_i and target confidence weight w^t_i in the MEA framework; no value, grid, or sensitivity analysis is provided, so reported gains may depend on this choice.
assumptions (4)
  • ad hoc to paper Predictive coding theory is an appropriate explanatory model for source-free adaptation.
    Section III-B maps predictive coding to SFDA by analogy; no formal mathematical link or causal mechanism is derived.
  • domain assumption Accuracy of a source model on other visible source domains is a reliable proxy for its relevance to the target domain.
    Used in Eq. (7) to set source model weights; can fail when visible source domains are unrelated to the target.
  • domain assumption Average softmax confidence on target samples reflects source model alignment with the target.
    Used in Eq. (8); softmax confidence is often miscalibrated and no calibration analysis is given.
  • domain assumption Baseline accuracies reported in earlier papers are comparable to the authors' reproduced runs.
    Fig. 2 caption states some results are 'recorded as per the original publication', which may use different splits or hyperparameters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unveiling the Superior Paradigm: A Comparative Study of Source-Free Domain Adaptation and Unsupervised Domain Adaptation." pith.science (2026). https://pith.science/paper/QSNLQRLV

@misc{pith2026241115844,
  author       = {Pith},
  title        = {Pith review of: Unveiling the Superior Paradigm: A Comparative Study of Source-Free Domain Adaptation and Unsupervised Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QSNLQRLV}},
  note         = {Machine review of arXiv:2411.15844}
}
read the original abstract

In domain adaptation, there are two popular paradigms: Unsupervised Domain Adaptation (UDA), which aligns distributions using source data, and Source-Free Domain Adaptation (SFDA), which leverages pre-trained source models without accessing source data. Evaluating the superiority of UDA versus SFDA is an open and timely question with significant implications for deploying adaptive algorithms in practical applications. In this study, we demonstrate through predictive coding theory and extensive experiments on multiple benchmark datasets that SFDA generally outperforms UDA in real-world scenarios. Specifically, SFDA offers advantages in time efficiency, storage requirements, targeted learning objectives, reduced risk of negative transfer, and increased robustness against overfitting. Notably, SFDA is particularly effective in mitigating negative transfer when there are substantial distribution discrepancies between source and target domains. Additionally, we introduce a novel data-model fusion scenario, where data sharing among stakeholders varies (e.g., some provide raw data while others provide only models), and reveal that traditional UDA and SFDA methods do not fully exploit their potential in this context. To address this limitation and capitalize on the strengths of SFDA, we propose a novel weight estimation method that effectively integrates available source data into multi-SFDA (MSFDA) approaches, thereby enhancing model performance within this scenario. This work provides a thorough analysis of UDA versus SFDA and advances a practical approach to model adaptation across diverse real-world environments.

Figures

Figures reproduced from arXiv: 2411.15844 by the authors.

Figure 1
Figure 1. Time analysis of UDA and SFDA methods in single-sourc [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Performance analysis with RN50. (a) and (b) depict si [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of SFDA and UDA learning objectives on the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Contrast between the learning objectives of SFDA met [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Analysis of negative transfer of SFDA methods and UDA [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Analysis of overfitting of SHOTAvg (SFDA method) on the office-Home and DomainNet datasets. The figure illustrates that training and test set accuracies are nearly identical, suggesting that source domain-independent adaptation algorithms based solely on the source dom…
Figure 7
Figure 7. Figure 7: The scenario of data-model fusion, where some source [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Model Estimation and Adaptation Framework for data [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 58 canonical work pages

  1. [1]

    Deep learning in static, metric-based bug prediction,

    R. Ferenc, D. B´ an, T. Gr´ osz, and T. Gyim ´ othy, “Deep learning in static, metric-based bug prediction,” Array, vol. 6, p. 100021, 2020

    work page 2020

  2. [2]

    A digital liquid stat e machine with biologically inspired learning and its application to spee ch recognition,

    Y . Zhang, P . Li, Y . Jin, and Y . Choe, “A digital liquid stat e machine with biologically inspired learning and its application to spee ch recognition,” IEEE transactions on neural networks and learning systems , vol. 26, no. 11, pp. 2635–2649, 2015

    work page 2015

  3. [3]

    Twitter mining for disa ster response: A domain adaptation approach.,

    H. Li, N. Guevara, N. Herndon, D. Caragea, K. Neppalli, C. Caragea, A. C. Squicciarini, and A. H. Tapia, “Twitter mining for disa ster response: A domain adaptation approach.,” in ISCRAM, 2015

    work page 2015

  4. [4]

    Conditional ad versarial domain adaptation,

    M. Long, Z. Cao, J. Wang, and M. I. Jordan, “Conditional ad versarial domain adaptation,” arXiv preprint arXiv:1705.10667 , 2017

    arXiv 2017

  5. [5]

    Learning transfe rable features with deep adaptation networks,

    M. Long, Y . Cao, J. Wang, and M. Jordan, “Learning transfe rable features with deep adaptation networks,” in International conference on machine learning, pp. 97–105, PMLR, 2015

    work page 2015

  6. [6]

    Analysis of representations for domain adaptation,

    S. Ben-David, J. Blitzer, K. Crammer, F. Pereira, et al. , “Analysis of representations for domain adaptation,” Advances in neural information processing systems, vol. 19, p. 137, 2007

    work page 2007

  7. [7]

    A theory of learning from different domains,

    S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pere ira, and J. W. V aughan, “A theory of learning from different domains, ” Machine learning, vol. 79, pp. 151–175, 2010

    work page 2010

  8. [8]

    Homm: Higher-order moment matching for unsupervised doma in adap- tation,

    C. Chen, Z. Fu, Z. Chen, S. Jin, Z. Cheng, X. Jin, and X.-S. H ua, “Homm: Higher-order moment matching for unsupervised doma in adap- tation,” in Proceedings of the AAAI conference on artificial intelligen ce, vol. 34, pp. 3422–3429, 2020

    work page 2020

Show all 66 references
  1. [9]

    M oment matching for multi-source domain adaptation,

    X. Peng, Q. Bai, X. Xia, Z. Huang, K. Saenko, and B. Wang, “M oment matching for multi-source domain adaptation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 1406– 1415, 2019

    2019

  2. [10]

    Robust unsupervised domain adapta tion for neural networks via moment alignment,

    W. Zellinger, B. A. Moser, T. Grubinger, E. Lughofer, T. Natschl¨ ager, and S. Saminger-Platz, “Robust unsupervised domain adapta tion for neural networks via moment alignment,” Information Sciences, vol. 483, pp. 174–191, 2019

    2019

  3. [11]

    A class-aware optimal transport approach with higher-ord er mo- ment matching for unsupervised domain adaptation,

    T. Nguyen, V . Nguyen, T. Le, H. Zhao, Q. H. Tran, and D. Phu ng, “A class-aware optimal transport approach with higher-ord er mo- ment matching for unsupervised domain adaptation,” arXiv preprint arXiv:2401.15952, 2024

    2024 arXiv

  4. [12]

    Robust loc al preserving and global aligning network for adversarial domain adaptat ion,

    W. Qiang, J. Li, C. Zheng, B. Su, and H. Xiong, “Robust loc al preserving and global aligning network for adversarial domain adaptat ion,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 3, pp. 3014–3029, 2021

    2021

  5. [13]

    Adversar ial dropout regularization,

    K. Saito, Y . Ushiku, T. Harada, and K. Saenko, “Adversar ial dropout regularization,” arXiv preprint arXiv:1711.01575 , 2017

    2017 arXiv

  6. [14]

    Advers arial discrim- inative domain adaptation,

    E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell, “Advers arial discrim- inative domain adaptation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 7167–7176, 2017

    2017

  7. [15]

    Unsupervised domain adapt ation via structurally regularized deep clustering,

    H. Tang, K. Chen, and K. Jia, “Unsupervised domain adapt ation via structurally regularized deep clustering,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni tion (CVPR), June 2020

    2020

  8. [16]

    Do we really need to access t he source data ? source hypothesis transfer for unsupervised domain a daptation,

    J. Liang, D. Hu, and J. Feng, “Do we really need to access t he source data ? source hypothesis transfer for unsupervised domain a daptation,” in International Conference on Machine Learning , pp. 6028–6039, PMLR, 2020

    2020

  9. [17]

    Source data- absent unsupervised domain adaptation through hypothesis transf er and labeling transfer,

    J. Liang, D. Hu, Y . Wang, R. He, and J. Feng, “Source data- absent unsupervised domain adaptation through hypothesis transf er and labeling transfer,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2021

    2021

  10. [18]

    Balancing discrimina bility and transferability for source-free domain adaptation,

    J. N. Kundu, A. R. Kulkarni, S. Bhambri, D. Mehta, S. A. Ku lkarni, V . Jampani, and V . B. Radhakrishnan, “Balancing discrimina bility and transferability for source-free domain adaptation,” in International Con- ference on Machine Learning , pp. 11710–11728, PMLR, 2022. 12

    2022

  11. [19]

    Source-fre e unsu- pervised domain adaptation with sample transport learning ,

    Q. Tian, C. Ma, F. Zhang, S. Peng, and H. Xue, “Source-fre e unsu- pervised domain adaptation with sample transport learning ,” J. Comput. Sci. Technol., vol. 36, no. 3, pp. 606–616, 2021

    2021

  12. [20]

    Exploiting the intrinsic neighborhood structure for source-free domain a daptation,

    S. Y ang, J. van de Weijer, L. Herranz, S. Jui, et al. , “Exploiting the intrinsic neighborhood structure for source-free domain a daptation,” Advances in Neural Information Processing Systems , vol. 34, 2021

    2021

  13. [21]

    General- ized source-free domain adaptation,

    S. Y ang, Y . Wang, J. van de Weijer, L. Herranz, and S. Jui, “General- ized source-free domain adaptation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 8978–8987, 2021

    2021

  14. [22]

    Attracting and dispersing: A simple approach for source-free domain adaptation,

    S. Y ang, S. Jui, J. van de Weijer, et al. , “Attracting and dispersing: A simple approach for source-free domain adaptation,” Advances in Neural Information Processing Systems , vol. 35, pp. 5802–5815, 2022

    2022

  15. [23]

    On e ring to bring them all: Towards open-set recognition under domain s hift,

    S. Y ang, Y . Wang, K. Wang, S. Jui, and J. van de Weijer, “On e ring to bring them all: Towards open-set recognition under domain s hift,” arXiv preprint arXiv: 2206.03600 , 2022

    2022 arXiv

  16. [24]

    Predictive coding in the vis ual cortex: a functional interpretation of some extra-classical recept ive-field effects,

    R. P . Rao and D. H. Ballard, “Predictive coding in the vis ual cortex: a functional interpretation of some extra-classical recept ive-field effects,” Nature neuroscience, vol. 2, no. 1, pp. 79–87, 1999

    1999

  17. [25]

    A review of predictive coding algorit hms,

    M. W. Spratling, “A review of predictive coding algorit hms,” Brain and cognition, vol. 112, pp. 92–97, 2017

    2017

  18. [26]

    Deep transfer learning with joint adaptation networks,

    M. Long, H. Zhu, J. Wang, and M. I. Jordan, “Deep transfer learning with joint adaptation networks,” in International conference on machine learning, pp. 2208–2217, PMLR, 2017

    2017

  19. [27]

    Bridging theor y and algo- rithm for domain adaptation,

    Y . Zhang, T. Liu, M. Long, and M. Jordan, “Bridging theor y and algo- rithm for domain adaptation,” in International Conference on Machine Learning, pp. 7404–7413, PMLR, 2019

    2019

  20. [28]

    Larger norm more transf erable: An adaptive feature norm approach for unsupervised domain ada ptation,

    R. Xu, G. Li, J. Y ang, and L. Lin, “Larger norm more transf erable: An adaptive feature norm approach for unsupervised domain ada ptation,” in Proceedings of the IEEE/CVF International Conference on Co mputer Vision, pp. 1426–1435, 2019

    2019

  21. [29]

    Unsupervised domain adapta tion by backpropagation,

    Y . Ganin and V . Lempitsky, “Unsupervised domain adapta tion by backpropagation,” in International conference on machine learning , pp. 1180–1189, PMLR, 2015

    2015

  22. [30]

    Deep cocktail n etwork: Multi-source unsupervised domain adaptation with categor y shift,

    R. Xu, Z. Chen, W. Zuo, J. Y an, and L. Lin, “Deep cocktail n etwork: Multi-source unsupervised domain adaptation with categor y shift,” in Proceedings of the IEEE conference on computer vision and pa ttern recognition, pp. 3964–3973, 2018

    2018

  23. [31]

    Aligning domain-specifi c distribution and classifier for cross-domain classification from multipl e sources,

    Y . Zhu, F. Zhuang, and D. Wang, “Aligning domain-specifi c distribution and classifier for cross-domain classification from multipl e sources,” in Proceedings of the AAAI conference on artificial intelligen ce, vol. 33, pp. 5989–5996, 2019

    2019

  24. [32]

    Stem: An approach to multi-source domain adaptation with guarant ees,

    V .-A. Nguyen, T. Nguyen, T. Le, Q. H. Tran, and D. Phung, “ Stem: An approach to multi-source domain adaptation with guarant ees,” in Proceedings of the IEEE/CVF International Conference on Co mputer Vision, pp. 9352–9363, 2021

    2021

  25. [33]

    Curr iculum manager for source selection in multi-source domain adaptation,

    L. Y ang, Y . Balaji, S.-N. Lim, and A. Shrivastava, “Curr iculum manager for source selection in multi-source domain adaptation,” i n Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, A ugust 23–28, 2020, Proceedings, Part XIV 16 , pp. 608–624, Springer, 2020

    2020

  26. [34]

    Information-theoretic regula rization for multi-source domain adaptation,

    G. Y . Park and S. W. Lee, “Information-theoretic regula rization for multi-source domain adaptation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 9214–9223, 2021

    2021

  27. [35]

    Y our classifier can secretly suffice multi-source domain adaptation,

    N. V enkat, J. N. Kundu, D. Singh, A. Revanur, et al. , “Y our classifier can secretly suffice multi-source domain adaptation,” Advances in Neural Information Processing Systems , vol. 33, pp. 4647–4659, 2020

    2020

  28. [36]

    Multi-source distilling domain ad aptation,

    S. Zhao, G. Wang, S. Zhang, Y . Gu, Y . Li, Z. Song, P . Xu, R. H u, H. Chai, and K. Keutzer, “Multi-source distilling domain ad aptation,” in Proceedings of the AAAI Conference on Artificial Intelligen ce, vol. 34, pp. 12975–12983, 2020

    2020

  29. [37]

    Multi-source contrib ution learning for domain adaptation,

    K. Li, J. Lu, H. Zuo, and G. Zhang, “Multi-source contrib ution learning for domain adaptation,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 10, pp. 5293–5307, 2021

    2021

  30. [38]

    Domain aggregat ion networks for multi-source domain adaptation,

    J. Wen, R. Greiner, and D. Schuurmans, “Domain aggregat ion networks for multi-source domain adaptation,” in International conference on machine learning, pp. 10214–10224, PMLR, 2020

    2020

  31. [39]

    Tmda: Task-specific m ulti-source domain adaptation via clustering embedded adversarial tra ining,

    H. Wang, W. Y ang, Z. Lin, and Y . Y u, “Tmda: Task-specific m ulti-source domain adaptation via clustering embedded adversarial tra ining,” in 2019 IEEE International Conference on Data Mining (ICDM) , pp. 1372– 1377, IEEE, 2019

    2019

  32. [40]

    Model adaptation: Historical contrastive learning for unsupervised domain adaptation w ithout source data,

    J. Huang, D. Guan, A. Xiao, and S. Lu, “Model adaptation: Historical contrastive learning for unsupervised domain adaptation w ithout source data,” Advances in Neural Information Processing Systems , vol. 34, 2021

    2021

  33. [41]

    Adaptive adversarial netw ork for source- free domain adaptation,

    H. Xia, H. Zhao, and Z. Ding, “Adaptive adversarial netw ork for source- free domain adaptation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 9010–9019, 2021

    2021

  34. [42]

    Model adapt ation: Unsupervised domain adaptation without source data,

    R. Li, Q. Jiao, W. Cao, H.-S. Wong, and S. Wu, “Model adapt ation: Unsupervised domain adaptation without source data,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco gnition, pp. 9641–9650, 2020

    2020

  35. [43]

    Do main im- pression: A source data free domain adaptation method,

    V . K. Kurmi, V . K. Subramanian, and V . P . Namboodiri, “Do main im- pression: A source data free domain adaptation method,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Comput er Vision, pp. 615–625, 2021

    2021

  36. [44]

    Domain adaptive vi deo segmentation via temporal consistency regularization,

    D. Guan, J. Huang, A. Xiao, and S. Lu, “Domain adaptive vi deo segmentation via temporal consistency regularization,” i n Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 8053– 8064, 2021

    2021

  37. [45]

    Alleviating s tyle sensitivity then adapting: Source-free domain adaptation for medical i mage seg- mentation,

    Y . Y e, Z. Liu, Y . Zhang, J. Li, and H. Shen, “Alleviating s tyle sensitivity then adapting: Source-free domain adaptation for medical i mage seg- mentation,” in Proceedings of the 30th ACM International Conference on Multimedia , pp. 1935–1944, 2022

    1935

  38. [46]

    Source free domain adaptation for medical image segmentation with fourier style mining,

    C. Y ang, X. Guo, Z. Chen, and Y . Y uan, “Source free domain adaptation for medical image segmentation with fourier style mining,” Medical Image Analysis , vol. 79, p. 102457, 2022

    2022

  39. [47]

    Unsupervised multi-source domain adaptation without access to source data,

    S. M. Ahmed, D. S. Raychaudhuri, S. Paul, S. Oymak, and A. K. Roy- Chowdhury, “Unsupervised multi-source domain adaptation without access to source data,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 10103–10112, 2021

    2021

  40. [48]

    Confident anc hor-induced multi-source free domain adaptation,

    J. Dong, Z. Fang, A. Liu, G. Sun, and T. Liu, “Confident anc hor-induced multi-source free domain adaptation,” Advances in Neural Information Processing Systems, vol. 34, pp. 2848–2860, 2021

    2021

  41. [49]

    Evidentia l multi- source-free unsupervised domain adaptation,

    J. Pei, A. Men, Y . Liu, X. Zhuang, and Q. Chen, “Evidentia l multi- source-free unsupervised domain adaptation,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024

    2024

  42. [50]

    On balancing bias and v ariance in unsupervised multi-source-free domain adaptation,

    M. Shen, Y . Bu, and G. W. Wornell, “On balancing bias and v ariance in unsupervised multi-source-free domain adaptation,” in International Conference on Machine Learning , pp. 30976–30991, PMLR, 2023

    2023

  43. [51]

    Target-discri minability- induced multi-source-free domain adaptation,

    G. Li, Q. Zhang, P . Wang, R. He, and C. Wu, “Target-discri minability- induced multi-source-free domain adaptation,” in 2023 IEEE Interna- tional Conference on Image Processing (ICIP) , pp. 76–80, IEEE, 2023

    2023

  44. [52]

    Discriminability and transferability estimation: a baye sian source im- portance estimation approach for multi-source-free domai n adaptation,

    Z. Han, Z. Zhang, F. Wang, R. He, W. Su, X. Xi, and Y . Yin, “Discriminability and transferability estimation: a baye sian source im- portance estimation approach for multi-source-free domai n adaptation,” in Proceedings of the AAAI Conference on Artificial Intelligen ce, vol. ...

    2023

  45. [53]

    Adapting visual category models to new domains,

    K. Saenko, B. Kulis, M. Fritz, and T. Darrell, “Adapting visual category models to new domains,” in European conference on computer vision , pp. 213–226, Springer, 2010

    2010

  46. [54]

    Deep hashing network for unsupervised domain adaptation,

    H. V enkateswara, J. Eusebio, S. Chakraborty, and S. Pan chanathan, “Deep hashing network for unsupervised domain adaptation, ” in Pro- ceedings of the IEEE conference on computer vision and patte rn recognition, pp. 5018–5027, 2017

    2017

  47. [55]

    Unbiased metric lear ning: On the utilization of multiple datasets and web images for softeni ng bias,

    C. Fang, Y . Xu, and D. N. Rockmore, “Unbiased metric lear ning: On the utilization of multiple datasets and web images for softeni ng bias,” in Proceedings of the IEEE International Conference on Comput er Vision, pp. 1657–1664, 2013

    2013

  48. [56]

    Recognition in ter ra incognita,

    S. Beery, G. V an Horn, and P . Perona, “Recognition in ter ra incognita,” in Proceedings of the European conference on computer vision ( ECCV), pp. 456–473, 2018

    2018

  49. [57]

    In search of lost domain generalization,

    I. Gulrajani and D. Lopez-Paz, “In search of lost domain generalization,” arXiv preprint arXiv:2007.01434 , 2020

    2007 arXiv

  50. [58]

    Deep residual learni ng for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learni ng for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 770–778, 2016

    2016

  51. [59]

    Domain-adversaria l training of neural networks,

    Y . Ganin, E. Ustinova, H. Ajakan, P . Germain, H. Laroche lle, F. Lavi- olette, M. Marchand, and V . Lempitsky, “Domain-adversaria l training of neural networks,” The journal of machine learning research , vol. 17, no. 1, pp. 2096–2030, 2016

    2016

  52. [60]

    Maximu m classifier discrepancy for unsupervised domain adaptation,

    K. Saito, K. Watanabe, Y . Ushiku, and T. Harada, “Maximu m classifier discrepancy for unsupervised domain adaptation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 3723– 3732, 2018

    2018

  53. [61]

    Minimum class conf usion for versatile domain adaptation,

    Y . Jin, X. Wang, M. Long, and J. Wang, “Minimum class conf usion for versatile domain adaptation,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proc eedings, Part XXI 16 , pp. 464–480, Springer, 2020

    2020

  54. [62]

    Unc ertainty- induced transferability representation for source-free u nsupervised do- main adaptation,

    J. Pei, Z. Jiang, A. Men, L. Chen, Y . Liu, and Q. Chen, “Unc ertainty- induced transferability representation for source-free u nsupervised do- main adaptation,” IEEE Transactions on Image Processing , vol. 32, pp. 2033–2048, 2023. 13

    2023

  55. [63]

    Source- free domain adaptation via avatar prototype generation and adaptation,

    Z. Qiu, Y . Zhang, H. Lin, S. Niu, Y . Liu, Q. Du, and M. Tan, “ Source- free domain adaptation via avatar prototype generation and adaptation,” arXiv preprint arXiv:2106.15326 , 2021

    2021 arXiv

  56. [64]

    Learning to combine: Knowl- edge aggregation for multi-source domain adaptation,

    H. Wang, M. Xu, B. Ni, and W. Zhang, “Learning to combine: Knowl- edge aggregation for multi-source domain adaptation,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, A ugust 23–28, 2020, Proceedings, Part VIII 16 , pp. 727–744, Springer, 2020

    2020

  57. [65]

    T-svdnet: Explor ing high- order prototypical correlations for multi-source domain a daptation,

    R. Li, X. Jia, J. He, S. Chen, and Q. Hu, “T-svdnet: Explor ing high- order prototypical correlations for multi-source domain a daptation,” in Proceedings of the IEEE/CVF International Conference on Co mputer Vision, pp. 9991–10000, 2021

    2021

  58. [66]

    Most: Multi-source domain adaptation via optimal transport for s tudent-teacher learning,

    T. Nguyen, T. Le, H. Zhao, Q. H. Tran, T. Nguyen, and D. Phu ng, “Most: Multi-source domain adaptation via optimal transport for s tudent-teacher learning,” in Uncertainty in Artificial Intelligence , pp. 225–235, PMLR, 2021

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.