Pith. sign in

REVIEW 4 major objections 4 minor 33 references

SinSim: Sinkhorn-Regularized SimCLR

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Adding a Sinkhorn-regularized optimal transport term to SimCLR's loss produces a more structured latent space and better classification accuracy than plain SimCLR on all four datasets tested.

desk verdict SinSim adds a Sinkhorn regularizer to SimCLR; the empirical idea is fine, but the theory is wrong and the experiments lack the support to carry it. read the letter →

arxiv 2502.10478 v1 pith:CKIUD6O6 submitted 2025-02-13 cs.LG cs.CVstat.ML

M.Hadi Sepanj , Paul Fiegth This is my paper
classification cs.LGcs.CVstat.ML
keywords self-supervisedlearningcontrastiveSimCLRSinkhornregularizationoptimaltransportWassersteindistancerepresentationlatentspacestructure
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

SimCLR learns representations by pulling two augmented views of the same image together and pushing different images apart, but it does not explicitly control how the latent space is spread out globally. SinSim adds an entropy-regularized optimal transport (Sinkhorn) loss on the intermediate features, alongside the standard contrastive loss on the final embeddings. The paper argues that this regularizer promotes a well-dispersed, geometry-aware latent space and prevents mode collapse, leading to better class separation and downstream accuracy. The reported experiments show SinSim outperforming SimCLR on every dataset tested (MNIST, CIFAR-10, CIFAR-100, STL-10) for both linear and non-linear classifiers, and matching or beating VICReg and Barlow Twins on MNIST and STL-10. If these results hold, practitioners get a simple, drop-in regularization term for SimCLR-style training that improves representations without changing the architecture.

What carries the argument

The machinery is the entropy-regularized Wasserstein (Sinkhorn) distance, $W_\lambda(P,Q)=\min_{\gamma\in\Pi(P,Q)}\langle\gamma,C\rangle-\lambda H(\gamma)$, applied to the uniform empirical distributions of the features of the two augmented views, with cost $C_{ij}=\|h^{(i)}_1-h^{(j)}_2\|^2$. The Sinkhorn-Knopp algorithm solves the regularized problem by alternating row and column scaling, giving a differentiable transport plan $\gamma^*$. The load-bearing property is the Lemma 1 lower bound: the diagonal coupling $\tilde\gamma_{ij}=\frac{1}{N}\delta_{ij}$ is admissible, so the regularized loss is bounded below by the average diagonal distance minus the entropy term; the paper interprets this as enforced alignment plus dispersion. The entropy term's strict positivity (Gibbs-form solution) is then used to argue that mode collapse is prevented.

What would settle it

A two-point check settles Lemma 1: let $P=\frac12(\delta_a+\delta_b)$ and $Q=\frac12(\delta_b+\delta_a)$. The diagonal coupling $\gamma_{11}=\gamma_{22}=\frac12$ costs $\|a-b\|^2$, while the off-diagonal coupling $\gamma_{12}=\gamma_{21}=\frac12$ costs $0$; the off-diagonal plan is cheaper, so the paper's claim that any deviation from the diagonal increases the quadratic cost is false.

Watch

Extended reading notes

Core claim

SinSim's claim is that adding an entropy-regularized optimal transport term to SimCLR's loss, evaluated on the intermediate features rather than the final embeddings, yields a more structured and better-dispersed latent space. The paper derives a lower bound showing that as the regularization strength $\lambda$ vanishes, the optimal transport plan collapses toward the identity pairing, so the Sinkhorn loss becomes a diagonal alignment cost plus an entropy penalty; this is what it means for the regularizer to be geometry-aware. Empirically, the paper reports that SinSim outperforms SimCLR on all four benchmark datasets in both linear and non-linear classification, achieves the highest accuracy on MNIST and STL-10 compared with SimCLR, VICReg, and Barlow Twins, and produces UMAP embeddings with visibly better class separation.

Load-bearing premise

The paper's explanation for why SinSim works assumes that matching each feature in the first view to its counterpart in the second view is always the cheapest possible pairing; if swapping features can make the pairing cheaper, that explanation breaks.

Editorial extensions

If this is right

  • SinSim's linear classification accuracy exceeds plain SimCLR's on all four datasets, with the largest gains on MNIST (+1.4 points) and STL-10 (+2.3 points), and it takes the top accuracy among the compared methods on those two datasets.
  • The advantage carries over to a two-layer MLP probe, where SinSim again outperforms SimCLR everywhere and reaches the best score on MNIST and STL-10.
  • UMAP projections of SinSim embeddings show tighter, more separated class clusters than SimCLR on MNIST and CIFAR-10, indicating that the Sinkhorn term reduces inter-class overlap.
  • Ablations on MNIST and CIFAR-10 show that accuracy rises with the Sinkhorn weight $\beta$ up to roughly 0.8–0.9, that about 40 Sinkhorn iterations are optimal, and that the entropy strength $\lambda$ is best near 0.05–0.06.

Reading between the lines

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

  • The paper does not test the Sinkhorn regularizer on other contrastive or self-distillation frameworks; a direct extension would be to wrap the same entropy-regularized transport penalty around the features in BYOL or MoCo, where the projection head already produces two views of the same image.
  • Because the regularizer is computed on intermediate features, its cost depends on the choice of marginal distributions; the paper fixes uniform marginals without ablating alternatives such as data-dependent marginals, which could change the induced geometry.
  • A testable extension of the diagonal-alignment reading is to compute the Sinkhorn loss after shuffling the order of the second view's features; if shuffling changes the learned representations, the identity pairing is genuinely doing work, and if it does not, the benefit is attributable to the entropy term alone.
  • The ablations are based on 10-epoch pretraining with a single run; multi-seed, longer-training experiments would clarify whether the reported $\beta$ and $\lambda$ trends are stable, and whether the gains persist on larger architectures such as ResNet-50.
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 SinSim, an extension of SimCLR that adds a Sinkhorn-regularized optimal transport loss on the intermediate representations h to encourage a globally structured latent space. The total training loss is L = L_contrastive + β L_Sinkhorn, where L_Sinkhorn is an entropy-regularized Wasserstein distance. The authors provide two lemmas as theoretical justification (Lemma 1 on diagonal alignment, Lemma 2 on mode-collapse prevention), report linear and non-linear classification results on MNIST, CIFAR-10, CIFAR-100, and STL-10 against SimCLR, VICReg, and Barlow Twins, and include ablations over β, λ, and Sinkhorn iterations together with UMAP visualizations.

Significance. If the empirical results were reproducible and the theoretical account were correct, SinSim would offer a simple, potentially useful regularizer for contrastive learning, and the paper would provide a meaningful bridge between optimal transport and self-supervised representation learning. Strengths include the explicit ablations of the introduced hyperparameters, a falsifiable comparison on standard benchmarks, and a clearly stated design rationale for regularizing h instead of z. However, the central theoretical lemma is mathematically false, the loss definition is internally inconsistent, and the reported performance gains are not supported by error bars, seeds, or code. As it stands, neither the theoretical nor the empirical contribution is established.

major comments (4)
  1. [Section III-E, Lemma 1 and Eq. (7)] The proof of Lemma 1 asserts that any deviation from the diagonal pairing increases the quadratic cost, and concludes that ⟨γ, C⟩ ≥ (1/N) Σ_i ∥z1_i − z2_i∥² for every coupling γ. This is false: the diagonal coupling is merely one feasible coupling, so its cost is an upper bound on the optimal transport cost, not a lower bound. A concrete counterexample is N=2 with z1 = [(0,0), (1,1)] and z2 = [(1,1), (0,0)]: the diagonal average cost is 1, while the off-diagonal coupling γ12 = γ21 = 1/2 gives cost 0. Hence inequality (7) is not established, and the stated conclusion that minimizing Wλ forces alignment by the diagonal cost is unsupported. Since the Introduction lists a theoretical justification as a key contribution, this is a load-bearing error.
  2. [Section III-B Eq. (4) vs. Section III-E Eq. (6)] The entropy regularization term has inconsistent signs across the paper. Eq. (4) defines L_Sinkhorn = min_γ ⟨γ,C⟩ + λH(γ), while Eq. (6) defines Wλ(P,Q) = min_γ ⟨γ,C⟩ − λH(γ), with λ > 0. Since H(γ) ≥ 0 for discrete distributions, these are different objectives. The total loss in Eq. (5) uses L_Sinkhorn, but Lemma 2 and the Gibbs form in Eq. (9) follow the minus-sign convention. The reader therefore cannot determine which objective was actually optimized or implemented. This must be reconciled.
  3. [Section IV-D, step 1 vs. Section III-C] Section III-C argues that Sinkhorn regularization is applied to the intermediate representations h, citing information preservation and computational efficiency, and Eq. (4) indeed uses h. However, the implementation in Section IV-D, step 1, defines the cost matrix as Cij = ∥z(i)_1 − z(j)_2∥², i.e., on the final embeddings z, not h. If the experiments used z, the stated rationale and the computational-efficiency claim are contradicted; if they used h, the implementation description is wrong. Either way, the actual objective of the experiments is not clearly specified.
  4. [Section IV-B, Tables I and II] The central empirical claim is reported as single accuracy numbers without error bars, number of seeds, or code release. Self-supervised pretraining of ResNet-18 on CIFAR/STL datasets is known to have non-negligible run-to-run variance, so the reported improvements over SimCLR (0.5–2.3 percentage points) cannot be distinguished from noise on the basis of the evidence provided. The paper should report mean and standard deviation over at least three independent runs, and should release code and configuration details to make the numbers verifiable.
minor comments (4)
  1. [Section IV-E, Figures 2–7] The ablation narrative is inconsistent with the figures: the text states that results 'unambiguously demonstrate an improvement in classification accuracy as β increases,' while the caption of Figure 3 reports 'significant fluctuations' for CIFAR-10. Additionally, the ablation finds the best λ around 0.06, but the main experiments use λ = 0.05 selected from validation; this discrepancy should be explained.
  2. [Figure 2 caption] The caption says 'SimSim Performance'; this should be 'SinSim Performance'.
  3. [Reference [11]] Reference [11] cites van der Maaten and Hinton's t-SNE paper, but the text uses it for UMAP. The appropriate citation for UMAP is McInnes et al., 2018.
  4. [Section IV-C] The phrase 'results very much consistent of those in Table I' contains a grammatical error and should be revised.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the empirical claim is benchmarked against external baselines and no load-bearing argument reduces to its own inputs; Lemma 1's flaw is a mathematical error, not circularity.

full rationale

The paper's central empirical claim (SinSim beats SimCLR on MNIST and STL-10 and is competitive with VICReg and Barlow Twins) is evaluated by frozen-encoder linear and MLP probes on standard test sets, i.e., against external benchmarks, not quantities fit from the same data. The Sinkhorn objective is defined independently of the contrastive loss (Eqs. 4-5), and the claimed benefit is an empirical comparison. Hyperparameters β and λ are tuned on validation performance (Section III-D), and the reported test accuracies in Tables I and II are not predictions derived from those validation fits; this is standard model selection, not a fitted input renamed as a prediction. The paper does cite prior work by the same authors ([12], [13], [15]), but these references appear only in related work and are not used to justify the central mechanism; the Sinkhorn construction is supported by standard, external optimal-transport references ([7], [8], [16]-[18]). The main theoretical step, Lemma 1 in Section III-E, contains an unsupported assertion that "any deviation from the diagonal pairing increases the quadratic cost," which is false for permuted representations. However, this is a mathematical error in the proof and a correctness risk, not a circular reduction: the claimed lower bound does not follow from the definition of Wλ by construction, and the paper's empirical outcome is not logically forced by its inputs. No circular step can therefore be exhibited under the required standard.

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

The central claim rests on two tuned hyperparameters (β, λ) and an unproven assertion about the effect of uniform-marginal Sinkhorn regularization on intermediate features. The theoretical justification contains a false assumption about diagonal couplings, so the paper does not provide an independent derivation of why the regularizer should help.

free parameters (4)
  • β (Sinkhorn weighting) = tuned on validation; optimal around 0.8-0.9 in ablations
    Controls the balance between NT-Xent and Sinkhorn loss; tuned per dataset on validation performance (Section III-D, Section IV-E).
  • λ (entropy regularization strength) = 0.05 (selected) / 0.06 (optimal in ablation)
    Controls smoothness of the transport plan; selected from validation, with ablations showing best near 0.06 and a sharp drop for large values (Section III-D, Figures 6-7).
  • Sinkhorn iterations = 40
    Chosen from ablation showing best performance at 40 iterations (Section III-D, Figures 4-5).
  • Temperature τ in NT-Xent = not reported
    A key hyperparameter of the contrastive loss; not stated anywhere, so replication is impossible without it.
assumptions (4)
  • domain assumption The Sinkhorn loss with uniform marginals over the batch enforces a well-dispersed, geometry-aware latent space.
    Assumed in Section III-C and III-E; no proof or ablation on the choice of marginals is given.
  • ad hoc to paper Deviations from the diagonal pairing increase the quadratic cost.
    Used in Lemma 1's proof (Section III-E); this is false for permuted point sets, where off-diagonal couplings can reduce cost.
  • standard math The Sinkhorn-Knopp algorithm converges to the optimal coupling.
    Invoked in Section III-D and Lemma 2; standard for entropic OT, but convergence rate and numerical stability are not discussed.
  • domain assumption Regularizing h rather than z preserves information and improves final embedding quality.
    Argued in Section III-C without formal proof or ablations comparing regularization on h vs. z.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SinSim: Sinkhorn-Regularized SimCLR." pith.science (2026). https://pith.science/paper/CKIUD6O6

@misc{pith2026250210478,
  author       = {Pith},
  title        = {Pith review of: SinSim: Sinkhorn-Regularized SimCLR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CKIUD6O6}},
  note         = {Machine review of arXiv:2502.10478}
}
read the original abstract

Self-supervised learning has revolutionized representation learning by eliminating the need for labeled data. Contrastive learning methods, such as SimCLR, maximize the agreement between augmented views of an image but lack explicit regularization to enforce a globally structured latent space. This limitation often leads to suboptimal generalization. We propose SinSim, a novel extension of SimCLR that integrates Sinkhorn regularization from optimal transport theory to enhance representation structure. The Sinkhorn loss, an entropy-regularized Wasserstein distance, encourages a well-dispersed and geometry-aware feature space, preserving discriminative power. Empirical evaluations on various datasets demonstrate that SinSim outperforms SimCLR and achieves competitive performance against prominent self-supervised methods such as VICReg and Barlow Twins. UMAP visualizations further reveal improved class separability and structured feature distributions. These results indicate that integrating optimal transport regularization into contrastive learning provides a principled and effective mechanism for learning robust, well-structured representations. Our findings open new directions for applying transport-based constraints in self-supervised learning frameworks.

Figures

Figures reproduced from arXiv: 2502.10478 by the authors.

Figure 1
Figure 1. UMAP visualization of learned embeddings. (a) and (c) show the embedding space for SimCLR, while (b) and (d) illustrate SinSim’s feature [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Effect of Sinkhorn Regularization on SimSim Performance, varying [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Effect of Sinkhorn iterations on SinSim classification accuracy on [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Effect of Sinkhorn iterations, as in Figure 4, but now assessed on [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Effect of Sinkhorn regularization strength [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Effect of Sinkhorn regularization strength [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 32 canonical work pages

  1. [8]

    Sinkhorn Distances: Lightspeed Computation of Optimal Transport,

    M. Cuturi, “Sinkhorn Distances: Lightspeed Computation of Optimal Transport,” in Proc. NeurIPS, 2013

    work page 2013

  2. [17]

    Sinkhorn Distances: Lightspeed Computation of Optimal Transport,

    M. Cuturi, “Sinkhorn Distances: Lightspeed Computation of Optimal Transport,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 26, 2013

    work page 2013

  3. [28]

    Unbalanced Optimal Transport in High-Dimensional Machine Learning,

    K. Fatras, R. Flamary, N. Courty, and A. Rakotomamonjy, “Unbalanced Optimal Transport in High-Dimensional Machine Learning,” in Adv. Neural Inf. Process. Syst. (NeurIPS) , 2021

    work page 2021

  4. [1]

    A Simple Framework for Contrastive Learning of Visual Representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A Simple Framework for Contrastive Learning of Visual Representations,” in Proc. Int. Conf. on Machine Learning (ICML) , 2020

    work page 2020

  5. [2]

    Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning,

    J.-B. Grill et al., “Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning,” in Proc. NeurIPS, 2020

    work page 2020

  6. [3]

    VICReg: Variance-Invariance- Covariance Regularization for Self-Supervised Learning,

    A. Bardes, J. Berthelot, and A. Arnoud, “VICReg: Variance-Invariance- Covariance Regularization for Self-Supervised Learning,” in Proc. NeurIPS, 2021

    work page 2021

  7. [4]

    Barlow Twins: Self-Supervised Learning via Redun- dancy Reduction,

    J. Zbontar et al., “Barlow Twins: Self-Supervised Learning via Redun- dancy Reduction,” in Proc. ICML, 2021

    work page 2021

  8. [5]

    Momentum Contrast for Unsupervised Visual Representation Learning,

    K. He, H. Fan, Y . Wu, S. Xie, and R. Girshick, “Momentum Contrast for Unsupervised Visual Representation Learning,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , pp. 9729–9738, 2020

    work page 2020

Show all 33 references
  1. [6]

    Loss Landscape of Self-Supervised Learning,

    L. Ziyin, E. S. Lubana, M. Ueda, and H. Tanaka, “Loss Landscape of Self-Supervised Learning,” in NeurIPS 2022 Workshop: Self-Supervised Learning-Theory and Practice , 2022

    2022

  2. [7]

    On the Transfer of Masses,

    L. V . Kantorovich, “On the Transfer of Masses,” 1942

    1942

  3. [9]

    Optimal Transport for Domain Adaptation,

    N. Courty, R. Flamary, D. Tuia, and A. Rakotomamonjy, “Optimal Transport for Domain Adaptation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 9, pp. 1853–1865, 2017

    2017

  4. [10]

    Learning with a Wasserstein Loss,

    C. Frogner et al., “Learning with a Wasserstein Loss,” in Proc. NeurIPS, 2015

    2015

  5. [11]

    Visualizing Data using t-SNE,

    L. van der Maaten and G. Hinton, “Visualizing Data using t-SNE,” J. Mach. Learn. Res. , vol. 9, pp. 2579–2605, 2008

    2008

  6. [12]

    Context-Aware Augmentation for Contrastive Self-Supervised Representation Learning,

    H. Sepanj and P. Fieguth, “Context-Aware Augmentation for Contrastive Self-Supervised Representation Learning,”J. Comput. Vis. Imaging Syst., vol. 9, no. 1, pp. 4–7, 2023

    2023

  7. [13]

    Aligning Feature Distributions in VICReg Using Maximum Mean Discrepancy for Enhanced Manifold Awareness in Self-Supervised Representation Learning,

    H. Sepanj and P. Fieguth, “Aligning Feature Distributions in VICReg Using Maximum Mean Discrepancy for Enhanced Manifold Awareness in Self-Supervised Representation Learning,” J. Comput. Vis. Imaging Syst., vol. 10, no. 1, pp. 13–18, 2024

    2024

  8. [14]

    Geometric Dataset Distances via Op- timal Transport,

    D. Alvarez-Melis and N. Fusi, “Geometric Dataset Distances via Op- timal Transport,” in Adv. Neural Inf. Process. Syst. (NeurIPS) , vol. 33, pp. 21428–21439, 2020

    2020

  9. [15]

    Self-Supervised Learning Using Nonlinear Dependence,

    M. H. Sepanj, B. Ghojogh, and P. Fieguth, “Self-Supervised Learning Using Nonlinear Dependence,” arXiv preprint arXiv:2501.18875 , 2025

    2025

  10. [16]

    Sinkhorn Distributionally Robust Opti- mization,

    J. Wang, R. Gao, and Y . Xie, “Sinkhorn Distributionally Robust Opti- mization,” in arXiv preprint arXiv:2109.11926 , 2021

    2021 arXiv

  11. [18]

    Optimal Transport: Old and New,

    C. Villani, “Optimal Transport: Old and New,” vol. 338, Springer, 2009

    2009

  12. [19]

    Joint Distribution Optimal Transportation for Domain Adaptation,

    N. Courty, R. Flamary, A. Habrard, and A. Rakotomamonjy, “Joint Distribution Optimal Transportation for Domain Adaptation,” in Adv. Neural Inf. Process. Syst. (NeurIPS) , 2017

    2017

  13. [20]

    Optimal Transport for Domain Adaptation,

    R. Flamary, N. Courty, and D. Tuia, “Optimal Transport for Domain Adaptation,” in IEEE Trans. Pattern Anal. Mach. Intell. , 2016

    2016

  14. [21]

    DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation,

    B. B. Damodaran, B. Kellenberger, R. Flamary, D. Tuia, and N. Courty, “DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation,” in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018

    2018

  15. [22]

    Wasserstein Generative Adversarial Networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein Generative Adversarial Networks,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2017

    2017

  16. [23]

    Wasserstein Auto-Encoders,

    I. Tolstikhin, O. Bousquet, S. Gelly, and B. Schoelkopf, “Wasserstein Auto-Encoders,” in Int. Conf. Learn. Represent. (ICLR) , 2018

    2018

  17. [24]

    Improved Training of Wasserstein GANs,

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. Courville, “Improved Training of Wasserstein GANs,” in Adv. Neural Inf. Process. Syst. (NeurIPS), 2017

    2017

  18. [25]

    Learning Generative Models with Sinkhorn Divergences,

    A. Genevay, G. Peyr ´e, and M. Cuturi, “Learning Generative Models with Sinkhorn Divergences,” in Proc. Int. Conf. Artif. Intell. Stat. (AISTATS) , 2019

    2019

  19. [27]

    Fast Computation of Wasserstein Barycen- ters,

    M. Cuturi and A. Doucet, “Fast Computation of Wasserstein Barycen- ters,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2014

    2014

  20. [29]

    Mapping Estimation for Discrete Optimal Transport,

    M. Perrot, N. Courty, R. Flamary, and A. Habrard, “Mapping Estimation for Discrete Optimal Transport,” in Adv. Neural Inf. Process. Syst. (NeurIPS), 2016

    2016

  21. [30]

    Wasserstein Dictionary Learning: Optimal Transport-based Unsupervised Nonlinear Dictionary Learning,

    J. Schmitz, R. Flamary, N. Courty, and A. Rakotomamonjy, “Wasserstein Dictionary Learning: Optimal Transport-based Unsupervised Nonlinear Dictionary Learning,” in IEEE Trans. Signal Process. , 2018

    2018

  22. [31]

    Sinkhorn Autoencoders,

    G. Patrini, R. Van den Berg, and P. Forre, “Sinkhorn Autoencoders,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2020

    2020

  23. [32]

    Learning Generative Models with Sinkhorn Divergences,

    A. Genevay, L. Chizat, F. Bach, M. Cuturi, and G. Peyr ´e, “Learning Generative Models with Sinkhorn Divergences,” in Proc. Conf. Uncer- tainty Artif. Intell. (UAI) , 2018

    2018

  24. [33]

    Self-labelling via Simul- taneous Clustering and Representation Learning,

    Y . M. Asano, C. Rupprecht, and A. Vedaldi, “Self-labelling via Simul- taneous Clustering and Representation Learning,” in Int. Conf. Learn. Represent. (ICLR), 2019

    2019

  25. [34]

    Contrastive Self-Supervised Learning: Review, Progress, Challenges and Future Research Direc- tions,

    P. Kumar, P. Rawat, and S. Chauhan, “Contrastive Self-Supervised Learning: Review, Progress, Challenges and Future Research Direc- tions,” in Int. J. Multimed. Inf. Retr. , vol. 11, no. 4, pp. 461–488, 2022

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.