Pith. sign in

REVIEW 6 major objections 5 minor 46 references

Domain Generalization via Pareto Optimal Gradient Matching

T0 review · 6 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read POGM claims a gradient-matching rule that keeps the ERM trajectory and reaches top benchmark accuracy at O(2K) cost.

desk verdict A plausible and cheap gradient-matching method whose headline update is written three inconsistent ways and whose key generalization link is asserted, not proved. read the letter →

arxiv 2507.14227 v1 pith:QEJ5UKRC submitted 2025-07-16 cs.LG cs.AI

Khoi Do , Duong Nguyen , Nam-Khanh Le , Quoc-Viet Pham , Binh-Son Hua , Won-Joo Hwang This is my paper · ORCID
classification cs.LGcs.AI
keywords domaingeneralizationgradientmatchinginnerproductParetooptimalitymeta-learninginvariantout-of-distribution
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

The paper targets gradient-based domain generalization, where a model is trained so that gradient directions agree across source domains. It claims that prior gradient-matching methods, which maximize gradient inner products or minimize gradient distances, actually produce fluctuating gradient directions and carry high computational overhead from second-order derivatives. To fix this, it proposes Pareto Optimality Gradient Matching (POGM), which collects per-domain gradient trajectories as data and, in a separate meta-update, maximizes the worst-case gradient inner product while staying inside a $\kappa$-hypersphere around the empirical risk minimization (ERM) gradient. The Pareto reduction yields a closed-form invariant gradient with cost $O(2 \times K)$ and no Hessian approximation. If the claims hold, POGM is a cheap gradient-matching method that beats ERM on the real multi-domain benchmarks in the paper (for example 88.4% on PACS and 82.0% on VLCS) while remaining competitive on synthetic ones.

What carries the argument

The machinery is a Pareto-front reduction applied to gradient vectors rather than losses. Per-domain gradient trajectories $h_i = \theta_i^{(E)} - \theta$ are treated as collected data; their mean is $h_{\mathrm{ERM}}$. Lemma 1 shows the average gradient inner product over domains is lower-bounded by the worst-case inner product, so the multi-objective maximization collapses to a max-min problem (Lemma 2). Imposing the constraint $\|h_{\mathrm{GIP-C}} - h_{\mathrm{ERM}}\|^2 \le \kappa \|h_{\mathrm{ERM}}\|^2$ and solving that max-min produces the closed-form invariant gradient of Theorem 1, which costs only $O(2K)$ instead of $O(K^2)$. A meta-learning wrapper separates the domain-wise updates from the invariant-gradient update, so no Hessian of the joint loss is ever formed.

What would settle it

Record $B_1 = \frac{1}{K^2}\sum_{i,j} D_{\mathrm{KL}}[p_i(y|x,\theta)\|p_j(y|x,\theta)]$ during POGM training on a dataset such as PACS: if $B_1$ does not systematically decrease while gradient inner products increase, the mechanism behind Theorem 3 is unsupported. Alternatively, construct a target domain whose gradient has lower inner product with every source-domain gradient than the source gradients have with each other (outside the convex hull by Lemma 4); if POGM still improves on that target, the theorem's premises are not the operative explanation.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 1: at each round $r$, the invariant gradient is $h^{(r)}_{\mathrm{GIP-C}} = h^{(r)}_{\mathrm{ERM}} + \frac{\kappa \|h^{(r)}_{\mathrm{ERM}}\|}{\|h^{(r)}_{\pi}\|} h^{(r)}_{\pi}$, where $h^{(r)}_{\mathrm{ERM}}$ is the average of the $K$ domain-specific gradient trajectories and $h^{(r)}_{\pi} = \sum_i \pi_i h^{(r)}_i$ is a weighted combination whose weights $\pi$ minimize $h^{(r)}_{\pi} \cdot h^{(r)}_{\mathrm{ERM}} + \sqrt{\kappa}\,\|h^{(r)}_{\mathrm{ERM}}\|\,\|h^{(r)}_{\pi}\|$. The minimization runs over the Pareto front, so the $O(K^2)$ sum over gradient inner-product pairs is replaced by the worst-case pair (Lemmas 1 and 2). The paper then argues that this update reduces gradient variance across domains, converges more stably than the Fishr baseline, and lowers the target risk bound when the target domain lies in the convex hull of the source domains (Theorem 3). Empirically, the method records the top average accuracy in the paper's comparisons, improving over ERM on VLCS, PACS, OfficeHome, Terra Incognita, and DomainNet.

Load-bearing premise

The load-bearing premise is that maximizing pairwise gradient inner products actually shrinks the divergence between what the source-domain predictors output (term $B_1$ in the risk bound) and that the target domain sits inside the convex hull of the source domains; the first step is asserted rather than proved, and if either fails the generalization theorem no longer follows.

Editorial extensions

If this is right

  • Gradient matching becomes a practical plug-in: the meta-update costs $O(2K)$, so it scales to many source domains without pairwise inner products or Hessian approximations.
  • Because the update stays near the ERM trajectory, POGM can be composed with representation-mixing methods; the paper reports 91.2% on PACS with SWAD and 71.4% on CMNIST with CIRL, versus 66.3% for POGM alone.
  • Per-domain gradient angles become positively correlated during training (correlations around 0.69 to 0.91 on the reported datasets), which the paper interprets as reduced gradient conflict and more stable convergence than Fishr's early rounds.
  • The theory identifies where the method should help most: targets inside the convex hull of source domains, and datasets where inter-domain predictor divergence (term B1) can be reduced by gradient matching.
  • Since only gradient trajectories are exchanged, the method is naturally suited to distributed settings where raw data cannot be shared.

Reading between the lines

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

  • A direct diagnostic follows from the theory: track $B_1 = \frac{1}{K^2}\sum_{i,j} D_{\mathrm{KL}}[p_i(y|x,\theta)\|p_j(y|x,\theta)]$ during POGM training; datasets where $B_1$ falls as gradient inner products rise should be the ones where POGM's gains over ERM are largest.
  • The same Pareto-weighted averaging with a trust-region constraint could transfer to multi-task learning or federated aggregation, where conflicting per-client gradients play the same role as per-domain gradients.
  • Because the weights $\pi$ are scalar, POGM can only rescale each domain gradient, not rotate it; a natural extension is per-coordinate or directional weights, which the paper itself flags as the next step and which would require a new closed-form solution.
  • For very similar source domains, Lemma 4 suggests the learned gradient stays inside the source convex hull and may miss a distant target, so combining POGM with test-time adaptation or stronger augmentation is the regime most likely to show additional gains.
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

6 major / 5 minor

Summary. The paper proposes Pareto Optimality Gradient Matching (POGM), a gradient-based domain generalization method. POGM maximizes pairwise gradient inner products across source domains while constraining the learned gradient to lie on a kappa-hypersphere around the ERM gradient trajectory. The method is implemented in a meta-learning framework to avoid second-order derivatives, and the main theoretical result (Theorem 1) claims a closed-form invariant gradient update of the form h_GIP-C = h_ERM + kappa*(||h_ERM||/||h_pi||)*h_pi, with pi minimizing h_pi . h_ERM + sqrt(kappa)*||h_ERM||*||h_pi||, at O(2K) cost. The empirical evaluation on DomainBed reports state-of-the-art or competitive results on PACS, VLCS, OfficeHome, Terra Incognita, and DomainNet, with ablations over the meta-learning rate, local epochs, and the hypersphere radius kappa.

Significance. If the theoretical derivation and the reported numbers are correct, POGM would be a valuable contribution: it is a simple, cheap gradient-matching method that avoids Hessian computations, achieves strong DomainBed results, and integrates well with representation-based methods. The paper also provides a useful empirical study of gradient fluctuation in Fish and Fishr, and the proposed invariant-gradient correlation analysis is a nice diagnostic. However, the theoretical support currently contains several serious and load-bearing inconsistencies, and the empirical reproducibility is compromised by unresolved discrepancies between the stated update rule, the appendix proof, the algorithm listing, and the tables. These issues must be resolved before the central claims can be accepted.

major comments (6)
  1. [Section 4.2, Eq. (5) and Appendix C.1, Theorem 4/Eq. (19) and proof] The central closed-form update is stated inconsistently. In Eq. (5), h_GIP-C uses a coefficient kappa multiplying (||h_ERM||/||h_pi||) h_pi, while the selector uses sqrt(kappa). In Appendix C.1, Theorem 4's statement (Eq. (19)) uses kappa in both places, but its proof obtains the stationarity condition h_GIP-C = h_ERM + h_pi/(2*gamma) (Eq. (24)) and the optimized gamma = ||h_pi||/(2*sqrt(kappa)*||h_ERM||) (Eq. (27)), which upon substitution gives h_GIP-C = h_ERM + sqrt(kappa)*(||h_ERM||/||h_pi||)*h_pi. Thus the statement, the appendix theorem, and the appendix proof disagree. This is not cosmetic: at kappa = 0.5, the two candidate updates differ by a factor of sqrt(0.5) ~ 0.707 in the deviation from the ERM gradient, and only the sqrt(kappa) form satisfies the constraint ||h_GIP-C - h_ERM||^2 = kappa*||h_ERM||^2 used to derive the relaxation. Since the manuscript does not specify which variant produced the results in Table 1 and Table 5, the headline numbers and the kappa-ablation are not interpretable.
  2. [Algorithm 1, lines 14-16] The algorithm introduces a third variant and leaves the actual update underspecified. Line 14 defines phi = kappa^2*||h_ERM||^2, which is never used anywhere in the algorithm or the paper. Line 16 states that theta is updated using h_GIP-C defined via Theorem 1, but the explicit formula for h_GIP-C is not written in the algorithm, so a reader cannot tell whether the implementation uses kappa or sqrt(kappa), or whether phi plays a role. This ambiguity is critical because the paper promises an open-source implementation but no code or repository link is provided anywhere in the manuscript, despite the Conclusion saying the experiments are 'reproducible with our open-source implementation.'
  3. [Section 4.3, Theorem 3, Eq. (6) and Appendix C.3] The generalization bound in Theorem 3 is not actually connected to the GIP objective. The proof in Appendix C.3 decomposes the target risk gap into two KL terms, B1 (predictive distributions) and B2 (input distributions), and then asserts in Eq. (37) that maximizing the pairwise gradient inner products reduces B1. No argument or reference is provided for this link. Lemma 3 only bounds the V-divergence between domains inside the convex hull of the source domains, and it does not mention gradients or GIP. Without a proof that GIP maximization decreases B1, the bound is not a theoretical guarantee for POGM; it is an unexplained assertion.
  4. [Section 4.3, Theorem 2 and Appendix C.2] Theorem 2 as stated is not a meaningful variance-reduction claim. The statement reads Var(U_i(theta^{(r+1,e)})) <= Var(U_i(theta^{(r+1,e)})), which is trivially true and is not what the surrounding text claims. The proof in Appendix C.2, Eq. (34), ends with Var(U_i(theta^{(r+1,e)})) <= Var(U_i(theta^{(r+1,e)})) / (E^*(eta^2 L / 2 - eta)), which is again not a bound unless the denominator is shown to be positive and greater than one, and the notation E^* is undefined. The intended variance reduction result is therefore not established.
  5. [Appendix B.5, proof of Lemma 2] The proof of Lemma 2 is circular. Eq. (12) writes Lavg(theta*) >= LPareto(theta*) = LPareto(theta*) and then concludes that theta* is a Pareto optimal solution. The key inequality Lavg(theta*) >= LPareto(theta*) is asserted rather than derived from the definitions of Pareto optimality and the max-min problem in Lemma 2. Since Lemma 2 is used to justify reducing the multi-objective GIP problem to the worst-case objective, this gap undermines the Pareto-optimality justification of the closed-form update.
  6. [Table 2 and Table 1] There is a direct empirical inconsistency between the two tables. Table 1 reports POGM's VLCS accuracy as 82.0 +/- 0.1, but Table 2's POGM row lists 70.0 +/- 0.3 in the VLCS column, which matches the OfficeHome value from Table 1 (70.0 +/- 0.3). This mislabeling makes the integrability results on VLCS (POGM + Mixup, + Data Aug., + CIRL, + SWAD) impossible to interpret. The authors must correct the table and clarify which dataset those numbers correspond to.
minor comments (5)
  1. [Conclusion and Appendix A.3] The Conclusion states that the experiments are reproducible with an open-source implementation, but no repository link or release information appears in the manuscript; the only link is the DomainBed repository in Appendix A.3. Please add a URL or state clearly that code will be released upon publication.
  2. [Equation (1)] The summation notation in Eq. (1) is ambiguous: it writes sum over i in K with i != j and j in K, but the index j is not bound properly. It should be written as sum over ordered or unordered pairs (i,j) with i != j.
  3. [Corollary 2] The inequality in Corollary 2 has the wrong direction if L denotes a loss to be minimized: it states L(theta*_GIP-C) > L(theta*_ERM) as evidence that GIP-C is 'better,' which would be true only if L were a utility to be maximized. Please correct the sign or clarify the convention.
  4. [References] There are duplicate reference entries: [8] and [9] are the same paper (EQRM), and [11] and [12] are the same paper (multi-task autoencoders). Please merge them and renumber.
  5. [Appendix C.2, Eq. (34)] The proof of Theorem 2 uses E^* without defining it, and the final inequality in Eq. (34) is dimensionally inconsistent (a variance bounded by a variance divided by a possibly negative quantity). Please rewrite the argument with clear notation and hypotheses.

Circularity Check

2 steps flagged · score 6.0 of 10

Circularity in the theory: Lemma 2's proof and Theorem 2's proof each reduce to their own conclusions; the headline update rule is also internally inconsistent (κ vs √κ), leaving the SOTA results as the main independent content.

  1. other [Appendix B.5 (Proof of Lemma 2), Eq. (12)]
    "≥ LPareto(θ∗) = LPareto(θ∗). (12) Therefore, we have that θ∗ is also the Pareto optimality solution."

    Lemma 2 is the load-bearing bridge that converts the multi-objective Pareto problem into the max-min problem (3) from which Theorem 1 is stated. Its proof assumes θ* is Pareto optimal (as in the lemma statement) and then 'derives' that conclusion from a chain ending in the tautology L_Pareto(θ*)=L_Pareto(θ*); no argument connects the max-min solution to Pareto optimality. Theorem 1's 'given the Pareto condition as mentioned in Lemma 2' therefore inherits an unproved, definitional equivalence.

  2. other [Appendix C.2 (Proof of Theorem 2), Eq. (34)]
    "Var(Ui(θ(r+1,e))) ≤ Uk(θ(r,E∗)) − Uk(θ(r,0)) / E∗(η2L/2 − η) = Var(Ui(θ(r+1,e))) / E∗(η2L/2 − η). (34)"

    The theorem claims the GIP variance decreases across meta-update rounds. In the proof, the target Var(U_i) is first bounded by a gradient-norm expression, and that expression is then equated to Var(U_i) itself divided by E*(η²L/2 − η). The claimed bound therefore reduces to Var ≤ Var / const, i.e., the quantity to be bounded is reused as its own bound. Since Theorem 2 is the theoretical guarantee for the method's invariant-gradient property, this self-referential inequality makes the guarantee circular rather than derived.

full rationale

The DomainBed benchmark results are external and are not circular evidence. The circularity is in the theoretical derivation chain: Lemma 2 (Appendix B.5, Eq. 12) proves Pareto optimality from the tautology L_Pareto(θ*)=L_Pareto(θ*), and Theorem 2 (Appendix C.2, Eq. 34) bounds Var by Var/E*(η²L/2 − η), reusing the target as its own bound. Additionally, the headline update is not even consistently derived: Theorem 1/Eq. (5) and Appendix Theorem 4 (Eq. 19) state the update coefficient as κ, while the proof's stationarity condition (Eq. 24) with the optimized γ (Eq. 27) yields √κ; Algorithm 1 line 14 defines an unused φ=κ²||h_ERM||². No code or repository link is provided despite the conclusion's promise of an open-source implementation, so which variant produced the Table 1 results is unverifiable. Theorem 3 (Eq. 6) also asserts without proof that maximizing GIP reduces the D_KL(y|x) term B1 in the risk bound. Because the empirical benchmark is independent, the circularity is partial rather than total, giving a score of 6.

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

The method introduces no new physical or conceptual entities; the kappa-hypersphere is a constraint, not an entity. It relies on tuned hyperparameters kappa, alpha, and E, and on the unproven assumption that GIP maximization decreases the label-conditional KL divergence between domains.

free parameters (4)
  • kappa (searching hypersphere radius) = 0.5 in main runs; ablation over 0.05, 0.1, 0.5
    Controls how far the learned gradient may deviate from the ERM trajectory; tuned per dataset via validation.
  • meta update learning rate alpha = 0.01 in main runs; ablation over 0.01, 0.1, 0.5
    Step size for the meta-update that applies the invariant gradient; tuned per dataset.
  • domain-specific iterations E = 5 in main runs; ablation over 1, 5, 10
    Number of local SGD steps per domain between meta-updates; affects the quality of gradient trajectory estimates.
  • inner SGD learning rate eta = 0.001 for CMNIST/RMNIST, 0.00005 for other datasets
    Learning rate for the domain-wise training updates; chosen per dataset following DomainBed defaults.
assumptions (4)
  • standard math The loss is L-smooth (Lipschitz gradient), used in Lemma 5 to bound gradient variance.
    Invoked in Appendix B.7 for the proof of Lemma 5 and Theorem 2; standard in optimization analysis, but the bound sign is mishandled.
  • domain assumption Target domain lies within the convex hull of source domains (V-divergence bound).
    Used in Lemma 3 and Theorem 3 to derive the target risk bound; not verified for DomainBed datasets.
  • ad hoc to paper Maximizing the gradient inner product sum over domain pairs reduces the hypothesis divergence term B1, the average pairwise KL divergence between per-domain predictors.
    Stated in Theorem 3 (eq. 6) without proof; this is the load-bearing step linking GIP to generalization, and no derivation is provided.
  • domain assumption Pareto optimality definitions and the max-min reduction (Lemma 1) apply to the gradient-matching objective.
    Used to justify replacing the sum of GIP pairs with worst-case optimization; the proof of Lemma 2 is circular.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Domain Generalization via Pareto Optimal Gradient Matching." pith.science (2026). https://pith.science/paper/QEJ5UKRC

@misc{pith2026250714227,
  author       = {Pith},
  title        = {Pith review of: Domain Generalization via Pareto Optimal Gradient Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QEJ5UKRC}},
  note         = {Machine review of arXiv:2507.14227}
}
read the original abstract

In this study, we address the gradient-based domain generalization problem, where predictors aim for consistent gradient directions across different domains. Existing methods have two main challenges. First, minimization of gradient empirical distance or gradient inner products (GIP) leads to gradient fluctuations among domains, thereby hindering straightforward learning. Second, the direct application of gradient learning to the joint loss function can incur high computation overheads due to second-order derivative approximation. To tackle these challenges, we propose a new Pareto Optimality Gradient Matching (POGM) method. In contrast to existing methods that add gradient matching as regularization, we leverage gradient trajectories as collected data and apply independent training at the meta-learner. In the meta-update, we maximize GIP while limiting the learned gradient from deviating too far from the empirical risk minimization gradient trajectory. By doing so, the aggregate gradient can incorporate knowledge from all domains without suffering gradient fluctuation towards any particular domain. Experimental evaluations on datasets from DomainBed demonstrate competitive results yielded by POGM against other baselines while achieving computational efficiency.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 44 canonical work pages

  1. [1]

    Albuquerque, J

    I. Albuquerque, J. Monteiro, M. Darvishi, T. H. Falk, and I. Mitliagkas. Generalizing to unseen domains via distribution matching.arXiv preprint arXiv:1911.00804, Mar. 2021

    arXiv 1911

  2. [2]

    Beery, G

    S. Beery, G. Van Horn, and P. Perona. Recognition in terra incognita. In Eur. Conf. Comput. Vis., Sep. 2018

    work page 2018

  3. [3]

    Blanchard, A

    G. Blanchard, A. A. Deshmukh, U. Dogan, G. Lee, and C. Scott. Domain generalization by marginal transfer learning. J. Mach. Learn. Res., Jan. 2021

    work page 2021

  4. [4]

    H. M. Bui, T. Tran, A. T. Tran, and D. Phung. Exploiting domain- specific features to enhance domain generalization. In Adv. Neural Inform. Process. Syst., 2021

    work page 2021

  5. [5]

    J. Cha, S. Chun, K. Lee, H.-C. Cho, S. Park, Y . Lee, and S. Park. SW AD: Domain generalization by seeking flat minima. In Adv. Neural Inform. Process. Syst., Dec. 2021

    work page 2021

  6. [6]

    L. Chen, Y . Zhang, Y . Song, Y . Shan, and L. Liu. Improved test-time adaptation for domain generalization. InIEEE Conf. Comput. Vis. Pattern Recog., Jun. 2023

    work page 2023

  7. [7]

    Dayal, V

    A. Dayal, V . K. B, L. R. Cenkeramaddi, C. K. Mohan, A. Kumar, and V . N. Balasubramanian. MADG: Margin-based adversarial learning for domain generalization. In Adv. Neural Inform. Process. Syst., Dec. 2023

    work page 2023

  8. [8]

    Eastwood, A

    C. Eastwood, A. Robey, S. Singh, J. V . Kügelgen, H. Hassani, G. J. Pappas, and B. Schölkopf. Probable domain generalization via quantile risk minimization. In Adv. Neural Inform. Process. Syst., Dec. 2022

    work page 2022

Show all 46 references
  1. [9]

    Eastwood, A

    C. Eastwood, A. Robey, S. Singh, J. von Kügelgen, H. Hassani, G. J. Pappas, and B. Schölkopf. Probable domain generalization via quantile risk minimization. In Adv. Neural Inform. Process. Syst., Dec. 2022

    2022

  2. [10]

    C. Finn, P. Abbeel, and S. Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In Int. Conf. Mach. Learn., Aug. 2017

    2017

  3. [12]

    Ghifary, W

    M. Ghifary, W. Kleijn, M. Zhang, and D. Balduzzi. Domain generaliza- tion for object recognition with multi-task autoencoders. In Int. Conf. Comput. Vis., Dec. 2015

    2015

  4. [13]

    Gulrajani and D

    I. Gulrajani and D. Lopez-Paz. In search of lost domain generalization. In Int. Conf. Learn. Represent., May 2021

    2021

  5. [14]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In IEEE Conf. Comput. Vis. Pattern Recog., Jun. 2016

    2016

  6. [15]

    Huang, H

    Z. Huang, H. Wang, E. P. Xing, and D. Huang. Self-challenging improves cross-domain generalization. In Eur. Conf. Comput. Vis., Aug. 2022

    2022

  7. [16]

    D. Kim, Y . Yoo, S. Park, J. Kim, and J. Lee. Selfreg: Self-supervised con- trastive regularization for domain generalization. In Int. Conf. Comput. Vis., Oct. 2021

    2021

  8. [17]

    Krueger, E

    D. Krueger, E. Caballero, J.-H. Jacobsen, A. Zhang, J. Binas, D. Zhang, R. L. Priol, and A. Courville. Out-of-distribution generalization via risk extrapolation (rex). In Int. Conf. Mach. Learn., Jul. 2021

    2021

  9. [18]

    D. Li, Y . Yang, Y . Song, and T. M. Hospedales. Deeper, broader and artier domain generalization. In Int. Conf. Comput. Vis., Oct. 2017

    2017

  10. [19]

    P. Li, D. Li, W. Li, S. Gong, Y . Fu, and T. M. Hospedales. A simple feature augmentation for domain generalization. In Int. Conf. Comput. Vis., Oct. 2021

    2021

  11. [20]

    Y . Li, X. Tian, M. Gong, Y . Liu, T. Liu, K. Zhang, and D. Tao. Deep domain generalization via conditional invariant adversarial networks. In Eur. Conf. Comput. Vis., Sep. 2018

    2018

  12. [21]

    F. Lv, J. Liang, S. Li, B. Zang, C. H. Liu, Z. Wang, and D. Liu. Causality inspired representation learning for domain generalization. In IEEE Conf. Comput. Vis. Pattern Recog., Jun. 2022

    2022

  13. [22]

    Mahajan, S

    D. Mahajan, S. Tople, and A. Sharma. Domain generalization using causal matching. In Int. Conf. Mach. Learn., Jul. 2021

    2021

  14. [23]

    H. Nam, H. Lee, J. Park, W. Yoon, and D. Yoo. Reducing domain gap by reducing style bias. In IEEE Conf. Comput. Vis. Pattern Recog., pages 8690–8699, Jun. 2021

    2021

  15. [24]

    A. T. Nguyen, T. Tran, Y . Gal, and A. G. Baydin. Domain invariant representation learning with domain density transformations. In Adv. Neural Inform. Process. Syst., Dec. 2021

    2021

  16. [25]

    A. T. Nguyen, T. Tran, Y . Gal, P. Torr, and A. G. Baydin. KL guided domain adaptation. In Int. Conf. Learn. Represent., 2022

    2022

  17. [26]

    Nguyen, K

    T. Nguyen, K. Do, B. Duong, and T. Nguyen. Domain generalisation via risk distribution matching. In Proc. Wint. Conf. on App. Com. Vis., Jan. 2024

    2024

  18. [27]

    Nichol, J

    A. Nichol, J. Achiam, and J. Schulman. On first-order meta-learning algorithms. arXiv preprint arXiv:1803.02999, Mar. 2018

    2018 arXiv

  19. [28]

    X. Peng, Q. Bai, X. Xia, Z. Huang, K. Saenko, and B. Wang. Moment matching for multi-source domain adaptation. In Int. Conf. Comput. Vis., Nov. 2019

    2019

  20. [29]

    A. Rame, C. Dancette, and M. Cord. Fishr: Invariant gradient variances for out-of-distribution generalization. In Int. Conf. Mach. Learn., Jul. 2022

    2022

  21. [30]

    Sagawa, P

    S. Sagawa, P. W. Koh, T. B. Hashimoto, and P. Liang. Distributionally robust neural networks. In Int. Conf. Learn. Represent., May 2020

    2020

  22. [31]

    Shahtalebi, J.-C

    S. Shahtalebi, J.-C. Gagnon-Audet, T. Laleh, M. Faramarzi, K. Ahuja, and I. Rish. Sand-mask: An enhanced gradient masking strategy for the discovery of invariances in domain generalization. In Eur. Conf. Comput. Vis., Oct. 2020

    2020

  23. [32]

    G. SHI, Q. Li, W. Zhang, J. Chen, and X.-M. Wu. Recon: Reducing conflicting gradients from the root for multi-task learning. In Int. Conf. Learn. Represent., Feb. 2023

    2023

  24. [33]

    Y . Shi, J. Seely, P. Torr, S. N, A. Hannun, N. Usunier, and G. Syn- naeve. Gradient matching for domain generalization. In Int. Conf. Learn. Represent., 2022

    2022

  25. [34]

    Y . Shu, Z. Cao, C. Wang, J. Wang, and M. Long. Open domain general- ization with domain-augmented meta-learning. In IEEE Conf. Comput. Vis. Pattern Recog., Jun. 2021

    2021

  26. [35]

    Sicilia, X

    A. Sicilia, X. Zhao, and S. J. Hwang. Domain adversarial neural networks for domain generalization: when it works and how to improve. Mach. Learn., Apr. 2023

    2023

  27. [36]

    Torralba and A

    A. Torralba and A. A. Efros. Unbiased look at dataset bias. In IEEE Conf. Comput. Vis. Pattern Recog., Oct. 2011

    2011

  28. [37]

    V . N. Vapnik. Statistical Learning Theory . Wiley-Interscience, Jun. 1998

    1998

  29. [38]

    Venkateswara, J

    H. Venkateswara, J. Eusebio, S. Chakraborty, and S. Panchanathan. Deep hashing network for unsupervised domain adaptation. In IEEE Conf. Comput. Vis. Pattern Recog., Jul. 2017

    2017

  30. [39]

    H. Wang, H. Si, B. Li, and H. Zhao. Provable domain generalization via invariant-feature subspace recovery. In Int. Conf. Mach. Learn., Jul. 2022

    2022

  31. [40]

    P. Wang, Z. Zhang, Z. Lei, and L. Zhang. Sharpness-aware gradient matching for domain generalization. In IEEE Conf. Comput. Vis. Pattern Recog., Jun. 2023

    2023

  32. [41]

    S. Yan, H. Song, N. Li, L. Zou, and L. Ren. Improve unsupervised domain adaptation with mixup training, Jan. 2020

    2020

  33. [42]

    H. Yao, Y . Wang, L. Zhang, J. Zou, and C. Finn. C-mixup: Improving generalization in regression. In Adv. Neural Inform. Process. Syst., Dec. 2022

    2022

  34. [43]

    Zhang, H

    M. Zhang, H. Marklund, N. Dhawan, A. Gupta, S. Levine, and C. Finn. Adaptive risk minimization: Learning to adapt to domain shift. In Adv. Neural Inform. Process. Syst., Dec. 2021

    2021

  35. [44]

    S. Zhao, M. Gong, T. Liu, H. Fu, and D. Tao. Domain generalization via entropy regularization. In Adv. Neural Inform. Process. Syst., Dec. 2020

    2020

  36. [45]

    K. Zhou, Y . Yang, Y . Qiao, and T. Xiang. Domain generalization with mixstyle. In Int. Conf. Learn. Represent., Dec. 2021

    2021

  37. [46]

    K. Zhou, Z. Liu, Y . Qiao, T. Xiang, and C. C. Loy. Domain generalization: A survey. IEEE Trans. Pattern Anal. Mach. Intell., Apr. 2023

    2023

  38. [47]

    Zitzler and L

    E. Zitzler and L. Thiele. Multiobjective evolutionary algorithms: a com- parative case study and the strength pareto approach. IEEE Transactions on Evolutionary Computation, 1999. A Experimental Settings A.1 Datasets Rotated MNIST [11] consists of 10000 digits in MNIST with di...

    1999

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.