pith. sign in

arxiv: 2606.27948 · v1 · pith:T7DBG47S · submitted 2026-06-26 · cs.LG

RECAST: Model Reconstruction via Counterfactual-Aware Wasserstein Geometry under Limited Data

Xuan Zhao , Lena Krieger , Zhuo Cao , Arya Bangun , Hanno Scharr , Ira Assent This is my paper

pith:T7DBG47Sreviewed 2026-06-29 05:14 UTCmodel grok-4.3open to challenge →

classification cs.LG
keywords counterfactual explanationsmodel reconstructionWasserstein geometrylimited datablack-box auditingfairness diagnosticssurrogate modelsdecision boundary
0
0 comments X

The pith

RECAST reconstructs black-box models from limited counterfactuals via Wasserstein barycentric prototypes without ongoing queries.

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

The paper develops RECAST to rebuild surrogate models of opaque machine learning systems using counterfactual explanations under tight data and access constraints. It frames counterfactuals as informative samples for both outcome classes inside a Wasserstein barycentric construction that corrects decision-boundary shifts. This yields stable high-fidelity surrogates even when samples are few and noisy, while supporting group fairness checks. A sympathetic reader would value the result because third-party auditing of deployed models becomes feasible without continuous platform access.

Core claim

RECAST builds behavioral surrogate models from counterfactual explanations by embedding them as samples for both classes inside Wasserstein barycentric prototypes. The construction counters decision boundary shifts that arise in low-sample regimes and delivers high surrogate fidelity without requiring online access to the target model during reconstruction. Experiments on real-world datasets confirm that the resulting surrogates remain accurate and stable under limited and noisy conditions while enabling systematic fairness diagnostics.

What carries the argument

Wasserstein barycentric prototypes that treat counterfactual explanations as informative samples for both classes to mitigate decision boundary shifts.

If this is right

  • High-fidelity surrogates can be obtained from far fewer samples than standard reconstruction methods require.
  • Reconstruction proceeds without repeated online queries to the target model after the initial data collection.
  • Group fairness diagnostics become available to auditors even when direct model access is restricted.
  • Performance remains stable when input data contains moderate noise levels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same barycentric construction could be tested with other local explanation types such as feature attributions to broaden reconstruction options.
  • The limited-access regime makes the approach relevant for auditing models behind privacy-preserving APIs.
  • One could measure whether the method scales to high-dimensional inputs by tracking fidelity decay as feature count grows.

Load-bearing premise

Counterfactual explanations can be treated as representative samples for both classes inside the Wasserstein barycentric construction without introducing systematic bias or overfitting.

What would settle it

A reconstruction trial on a dataset with known severe class imbalance in which the surrogate's accuracy on held-out points falls well below the original model's accuracy.

Figures

Figures reproduced from arXiv: 2606.27948 by Arya Bangun, Hanno Scharr, Ira Assent, Lena Krieger, Xuan Zhao, Zhuo Cao.

Figure 1
Figure 1. Figure 1: CFs (x) are generated for class 0 (blue) samples - minimal changes that flip the target’s prediction from class 0 → class 1 (pink). Naively treating these as class 1 training samples shifts the surrogate’s boundary (solid line) toward class 0, causing class 0 samples to be misclassified as class 1. This is overconfidence: the surrogate becomes overly certain about class 1 in regions where the target (dashe… view at source ↗
Figure 2
Figure 2. Figure 2: Kernel density estimates of target model scores yˆm(x) for both class inputs, CFs; common CF methods. Concentration well above decision threshold 0.5 means CFs overshoot the boundary, confirming RECAST’s approach of avoiding near-boundary assumptions. structure relevant to the target’s decisions. This aggregation effectively increases the usable signal per query, mitigating overfitting in low-query regimes… view at source ↗
Figure 3
Figure 3. Figure 3: CF-consistent Wasserstein lens. All points in the figure correspond to probability distributions in Wasserstein space. Each class c is associated with a CF-consistent feasible set Cc = BW2 (Pc, εc) ∩ BW2 (Pcf, δc), given by the intersection (colored face) of a Wasserstein ball centered at Pc (dashed circle) and a ball centered at Pcf (dotted circle). Although CFs are one-sided, Pcf acts as shared geometric… view at source ↗
Figure 4
Figure 4. Figure 4: Top: Full original dataset and decision tree classification (blue and pink); CFs as x (orange); points available for reconstruc￾tion circled (black). Bottom: Decision region between barycenters: Q0 (blue), Q1 (pink); black circles midpoints along the optimal transport path (grey lines), indicating the decision region Q0.5. carry then no influence on Q⋆ 0 , and the formulation reduces to standard binary cla… view at source ↗
Figure 5
Figure 5. Figure 5: Fidelity (mean ± std) of surrogate models on real-world datasets under varying query sizes. RECAST with superior performance. Ablation Study. We conduct an ablation study on differ￾ent prototype constructions, distance metrics, and CFs inte￾gration strategies in RECAST with all details in App. C.1. 4.2. Reconstruction Fidelity [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Absolute fairness (DTIDP (solid) and DTIEO (dashed)) differences between reconstructed and target model as a function of noise level τ . Lower values mean more faithful preservation of target model fairness characteristics. calibrated via classical WRO methods, but this lies beyond the scope of the low-query regime studied here. In privacy-sensitive auditing contexts, where datasets are small, access is re… view at source ↗
Figure 8
Figure 8. Figure 8: Loss convergence over epochs on Adult dataset. Repetition All quantitative analyses were averaged across 10 independent repetitions to account for variability in sampling and model behavior. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Crossover between CF-based and no-CF reconstruction on the Adult dataset. CF-based reconstruction achieves higher fidelity in low-data regimes, while the no-CF decision tree surrogate improves with increased natural samples and surpasses the CF-based method around n ≈ 1000 per class. C.6. Data Modalities / High-Dimensional Data Our framework is generally applicable to any data modality that can be embedded… view at source ↗
Figure 10
Figure 10. Figure 10: Fidelity results for various sample sizes and binary classification problems within the MNIST dataset. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Runtime for RECAST across different datasets and configurations. Each RECAST model is trained for 500 epochs. Task and target model. We consider a toxicity classification task (Jigsaw, 2018). A binary classifier m is trained on embeddings ψ(x) obtained from a standard toxicity dataset and is treated as a black-box model during reconstruction. One-sided textual CFs. To emulate one-sided recourse, CFs are g… view at source ↗
read the original abstract

Counterfactual explanations (CFs) help understand machine learning models by identifying minimal input changes that would lead to alternative model outcomes. Recent work demonstrates their utility for reconstructing black-box models, enabling third-party auditing of opaque decision systems for fairness and accountability. Still, CF-based reconstruction may suffer from decision boundary shifts, overfitting, and restrictive assumptions requiring online query access to target platforms. We propose REconstruction via Counterfactual-Aware waSserstein opTimization (RECAST) under limited data and restricted access, a behavioral surrogate model based on Wasserstein barycentric prototypes. Our approach addresses decision boundary shifts by incorporating CFs as informative, though less representative, samples for both classes, maintaining high surrogate fidelity in low-sample regimes without requiring online access during reconstruction. To enhance fairness auditing, our method enables systematic group fairness diagnostics. Experiments on real-world datasets and various setups show that RECAST effectively achieves high fidelity and query efficiency, as well as stable results even when the access is limited and noisy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes RECAST, a behavioral surrogate model for black-box model reconstruction under limited data and restricted access. It uses counterfactual explanations (CFs) as informative samples for both classes inside a Wasserstein barycentric prototype construction to address decision boundary shifts, claims high surrogate fidelity without requiring online access during reconstruction, supports systematic group fairness diagnostics, and reports strong experimental results on real-world datasets for fidelity, query efficiency, and stability under limited/noisy access.

Significance. If the central construction is shown to avoid bias from treating boundary-near CFs as class-conditional samples, the work could meaningfully advance practical auditing of opaque models in data-scarce regimes. The focus on Wasserstein geometry for limited-data surrogates and the explicit fairness-auditing application are timely strengths.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'incorporating CFs as informative, though less representative, samples for both classes' maintains high surrogate fidelity rests on the unshown assumption that the Wasserstein barycentric prototype construction corrects for the fact that CFs are deterministically generated near the decision surface rather than sampled from the class-conditional measure. No reweighting, regularization, or convergence argument is visible that would prevent the barycenter from depending on the CF generator.
  2. [Abstract] Abstract: under the stated regime of 'limited and noisy access,' the method must demonstrate that surrogate fidelity is not an artifact of the CF distribution. The skeptic concern that CFs are not exchangeable with ordinary samples is load-bearing for the reconstruction claim and requires an explicit correction term or empirical validation protocol (e.g., comparison against a baseline that excludes CFs from the barycenter).
minor comments (1)
  1. [Abstract] The abstract would benefit from a one-sentence statement of the precise optimization objective or barycentric weighting rule.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We address each major comment below and will incorporate clarifications and additional validation in the revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'incorporating CFs as informative, though less representative, samples for both classes' maintains high surrogate fidelity rests on the unshown assumption that the Wasserstein barycentric prototype construction corrects for the fact that CFs are deterministically generated near the decision surface rather than sampled from the class-conditional measure. No reweighting, regularization, or convergence argument is visible that would prevent the barycenter from depending on the CF generator.

    Authors: We agree that the abstract does not make the correction mechanism explicit. The full manuscript (Section 3) constructs the Wasserstein barycenter from both limited samples and CFs via optimal transport, which by design averages across the two classes and thereby reduces sensitivity to CFs being boundary-localized; the geometry itself supplies the robustness without an added reweighting term. To address the referee's point directly, we will revise the abstract to reference this property and add a short clarifying paragraph in the method section on why the barycenter does not collapse to the CF generator. This change will be made in the next version. revision: yes

  2. Referee: [Abstract] Abstract: under the stated regime of 'limited and noisy access,' the method must demonstrate that surrogate fidelity is not an artifact of the CF distribution. The skeptic concern that CFs are not exchangeable with ordinary samples is load-bearing for the reconstruction claim and requires an explicit correction term or empirical validation protocol (e.g., comparison against a baseline that excludes CFs from the barycenter).

    Authors: The experiments already compare RECAST against limited-data baselines without CFs and report improved fidelity when CFs are included, supporting that the gains are not merely an artifact. However, we acknowledge that a dedicated ablation isolating the barycenter construction with versus without CFs would strengthen the claim. We will add this baseline experiment to the revised manuscript, following the referee's suggestion, to provide the requested empirical validation protocol under limited and noisy access. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and description contain no equations, derivations, or explicit self-citations that could be inspected for reduction to inputs by construction. The core claim describes an approach that incorporates CFs into a Wasserstein barycentric construction to address boundary shifts, but presents this as a methodological choice rather than a fitted parameter renamed as prediction or a self-definitional loop. No load-bearing uniqueness theorem, ansatz smuggled via citation, or renaming of known results is visible. The derivation chain cannot be walked because no formal steps are shown; the paper therefore remains self-contained against external benchmarks on the basis of the given text.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no information on free parameters, background axioms, or newly postulated entities.

pith-pipeline@v0.9.1-grok · 5714 in / 1068 out tokens · 23860 ms · 2026-06-29T05:14:48.244516+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

95 extracted references · 12 canonical work pages

  1. [1]

    Barycenters in the Wasserstein Space , url =

    Martial Agueh and Guillaume Carlier , title =. 2011 , url =. doi:10.1137/100805741 , timestamp =

    work page doi:10.1137/100805741 2011

  2. [2]

    Model extraction from counterfactual explanations , journal =

    Ulrich A. Model extraction from counterfactual explanations , journal =

  3. [3]

    Characterizing the risk of fairwashing , booktitle =

    Ulrich A. Characterizing the risk of fairwashing , booktitle =. 2021 , url =

    2021

  4. [4]

    Blanchet and Yang Kang and Karthyek Rajhaa A

    Jose H. Blanchet and Yang Kang and Karthyek Rajhaa A. M. , title =. J. Appl. Probab. , volume =. 2019 , url =. doi:10.1017/JPR.2019.49 , timestamp =

    work page doi:10.1017/jpr.2019.49 2019

  5. [5]

    Feder Cooper and Katherine Lee and Matthew Jagielski and Milad Nasr and Arthur Conmy and Eric Wallace and David Rolnick and Florian Tram

    Nicholas Carlini and Daniel Paleka and Krishnamurthy Dj Dvijotham and Thomas Steinke and Jonathan Hayase and A. Feder Cooper and Katherine Lee and Matthew Jagielski and Milad Nasr and Arthur Conmy and Eric Wallace and David Rolnick and Florian Tram. Stealing part of a production language model , booktitle =. 2024 , url =

    2024

  6. [6]

    Mingliang Chen and Min Wu , title =

  7. [7]

    Faster Wasserstein Distance Estimation with the Sinkhorn Divergence , booktitle =

    L. Faster Wasserstein Distance Estimation with the Sinkhorn Divergence , booktitle =. 2020 , url =

    2020

  8. [8]

    2012 , url =

    Li Deng , title =. 2012 , url =. doi:10.1109/MSP.2012.2211477 , timestamp =

    work page doi:10.1109/msp.2012.2211477 2012

  9. [9]

    Model Reconstruction Using Counterfactual Explanations:

    Pasan Dissanayake and Sanghamitra Dutta , editor =. Model Reconstruction Using Counterfactual Explanations:. Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024 , year =

    2024

  10. [10]

    Trained Random Forests Completely Reveal your Dataset , booktitle =

    Julien Ferry and Ricardo Fukasawa and Timoth. Trained Random Forests Completely Reveal your Dataset , booktitle =. 2024 , url =

    2024

  11. [11]

    Interpolating between Optimal Transport and

    Jean Feydy and Thibault S. Interpolating between Optimal Transport and. The 22nd International Conference on Artificial Intelligence and Statistics,. 2019 , url =

    2019

  12. [12]

    Kleywegt , title =

    Rui Gao and Anton J. Kleywegt , title =. Math. Oper. Res. , volume =. 2023 , url =. doi:10.1287/MOOR.2022.1275 , timestamp =

    work page doi:10.1287/moor.2022.1275 2023

  13. [13]

    2020 , url =

    Xueluan Gong and Qian Wang and Yanjiao Chen and Wang Yang and Xinchang Jiang , title =. 2020 , url =. doi:10.1109/MCOM.001.2000196 , timestamp =

    work page doi:10.1109/mcom.001.2000196 2020

  14. [14]

    CoRR , volume =

    Awa Khouna and Julien Ferry and Thibaut Vidal , title =. CoRR , volume =. 2025 , url =. doi:10.48550/ARXIV.2502.05325 , eprinttype =. 2502.05325 , timestamp =

    work page doi:10.48550/arxiv.2502.05325 2025

  15. [15]

    ASIA CCS ’24, pp

    Jiacheng Liang and Ren Pang and Changjiang Li and Ting Wang , editor =. Model Extraction Attacks Revisited , booktitle =. 2024 , url =. doi:10.1145/3634737.3657002 , timestamp =

    work page doi:10.1145/3634737.3657002 2024

  16. [16]

    Advances in mathematics , volume=

    A convexity principle for interacting gases , author=. Advances in mathematics , volume=. 1997 , publisher=

    1997

  17. [17]

    Mathematical Programming , volume=

    Data-driven distributionally robust optimization using the Wasserstein metric: Performance guarantees and tractable reformulations , author=. Mathematical Programming , volume=. 2018 , publisher=

    2018

  18. [18]

    FAccT , pages =

    Ramaravind Kommiya Mothilal and Amit Sharma and Chenhao Tan , title =. FAccT , pages =

  19. [19]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Knockoff nets: Stealing functionality of black-box models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  20. [20]

    Martin Pawelczyk and Klaus Broelemann and Gjergji Kasneci , title =

  21. [21]

    Reza Shokri and Martin Strobel and Yair Zick , title =

  22. [22]

    Stealing Machine Learning Models via Prediction APIs , booktitle =

    Florian Tram. Stealing Machine Learning Models via Prediction APIs , booktitle =

  23. [23]

    NeurIPS , pages =

    Sohini Upadhyay and Shalmali Joshi and Himabindu Lakkaraju , title =. NeurIPS , pages =

  24. [24]

    Mittelstadt and Chris Russell , title =

    Sandra Wachter and Brent D. Mittelstadt and Chris Russell , title =. CoRR , volume =

  25. [25]

    FAccT , pages =

    Yongjie Wang and Hangwei Qian and Chunyan Miao , title =. FAccT , pages =

  26. [26]

    Lei You and Lele Cao and Mattias Nilsson and Bo Zhao and Lei Lei , title =

  27. [27]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    A survey on model extraction attacks and defenses for large language models , author=. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2 , pages=

  28. [28]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Activethief: Model extraction using active learning and unannotated public data , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  29. [29]

    29th USENIX security symposium (USENIX Security 20) , pages=

    High accuracy and high fidelity extraction of neural networks , author=. 29th USENIX security symposium (USENIX Security 20) , pages=

  30. [30]

    2018 , url =

    FICO , title =. 2018 , url =

    2018

  31. [31]

    Proceedings of the 3rd innovations in theoretical computer science conference , pages=

    Fairness through awareness , author=. Proceedings of the 3rd innovations in theoretical computer science conference , pages=

  32. [32]

    proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining , pages=

    Certifying and removing disparate impact , author=. proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining , pages=

  33. [33]

    Advances in neural information processing systems , volume=

    Equality of opportunity in supervised learning , author=. Advances in neural information processing systems , volume=

  34. [34]

    2016 , howpublished =

    Julia Angwin and Jeff Larson and Surya Mattu and Lauren Kirchner , title =. 2016 , howpublished =

    2016

  35. [35]

    1996 , howpublished =

    Adult Data Set , author =. 1996 , howpublished =

    1996

  36. [36]

    Kelley and Barry, Ronald , title =

    Pace, R. Kelley and Barry, Ronald , title =. 1997 , journal =

    1997

  37. [37]

    Proceedings of the 2017 ACM on Asia conference on computer and communications security , pages=

    Practical black-box attacks against machine learning , author=. Proceedings of the 2017 ACM on Asia conference on computer and communications security , pages=

    2017

  38. [38]

    Langley , title =

    P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

    2000

  39. [39]

    T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

    1980

  40. [40]

    M. J. Kearns , title =

  41. [41]

    Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

    1983

  42. [42]

    R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

    2000

  43. [43]

    Suppressed for Anonymity , author=

  44. [44]

    Newell and P

    A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981

    1981

  45. [45]

    A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959

    1959

  46. [46]

    Regulation (EU) 2024/1689 of the European Parliament and of the Council on laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) , year =

    2024

  47. [47]

    SIAM Journal on Scientific Computing , volume=

    Iterative Bregman projections for regularized transportation problems , author=. SIAM Journal on Scientific Computing , volume=. 2015 , publisher=

    2015

  48. [48]

    2008 , publisher=

    Optimal transport: old and new , author=. 2008 , publisher=

    2008

  49. [49]

    An Invitation to Statistics in Wasserstein Space , pages=

    The wasserstein space , author=. An Invitation to Statistics in Wasserstein Space , pages=. 2020 , publisher=

    2020

  50. [50]

    La Matematica , volume=

    A duality-based proof of the triangle inequality for the wasserstein distances , author=. La Matematica , volume=. 2024 , publisher=

    2024

  51. [51]

    POT: Python Optimal Transport , journal =

    R. POT: Python Optimal Transport , journal =. 2021 , volume =

    2021

  52. [52]

    Data Min

    Riccardo Guidotti , title =. Data Min. Knowl. Discov. , volume =

  53. [53]

    Selbst and Manish Raghavan , title =

    Solon Barocas and Andrew D. Selbst and Manish Raghavan , title =. FAT* , pages =

  54. [54]

    Sahil Verma and Varich Boonsanong and Minh Hoang and Keegan Hines and John Dickerson and Chirag Shah , title =

  55. [55]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Interpretable counterfactual explanations guided by prototypes , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  56. [56]

    Proceedings of the International Astronomical Union , volume=

    Prototype-based Models for the Supervised Learning of Classification Schemes , author=. Proceedings of the International Astronomical Union , volume=. 2016 , publisher=

    2016

  57. [57]

    Advances in Neural Information Processing Systems , volume=

    Prototypical Networks for Few-shot Learning , author=. Advances in Neural Information Processing Systems , volume=

  58. [58]

    Gloria Zen and Elisa Ricci , title =

  59. [59]

    2016 , url =

    I-Cheng Yeh , title =. 2016 , url =

    2016

  60. [60]

    arXiv preprint arXiv:2505.08847 , year=

    On the interplay of Explainability, Privacy and Predictive Performance with Explanation-assisted Model Extraction , author=. arXiv preprint arXiv:2505.08847 , year=

    work page arXiv

  61. [61]

    2017 IEEE International symposium on technologies for homeland security (HST) , pages=

    How to steal a machine learning classifier with deep learning , author=. 2017 IEEE International symposium on technologies for homeland security (HST) , pages=. 2017 , organization=

    2017

  62. [62]

    , author=

    InverseNet: Augmenting Model Extraction Attacks with Training Data Inversion. , author=. IJCAI , pages=

  63. [63]

    Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency , pages=

    Dualcf: Efficient model extraction attack from counterfactual explanations , author=. Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency , pages=

    2022

  64. [64]

    2009 IEEE 12th international conference on computer vision , pages=

    Fast and robust earth mover's distances , author=. 2009 IEEE 12th international conference on computer vision , pages=. 2009 , organization=

    2009

  65. [65]

    2009 , publisher =

    Optimal Transport: Old and New , author =. 2009 , publisher =

    2009

  66. [66]

    Now Publishers, February 2019

    Computational Optimal Transport , author =. Foundations and Trends in Machine Learning , volume =. 2019 , publisher =. doi:10.1561/2200000073 , issn =

    work page doi:10.1561/2200000073 2019

  67. [67]

    Entropy , volume=

    On wasserstein two-sample testing and related families of nonparametric tests , author=. Entropy , volume=. 2017 , publisher=

    2017

  68. [68]

    S´ ejourn´ e, J

    Sinkhorn divergences for unbalanced optimal transport , author=. arXiv preprint arXiv:1910.12958 , year=

    work page arXiv 1910

  69. [69]

    Advances in Neural Information Processing Systems , volume=

    Differential properties of sinkhorn approximation for learning with wasserstein distance , author=. Advances in Neural Information Processing Systems , volume=

  70. [70]

    Advances in neural information processing systems , volume=

    Stochastic optimization for large-scale optimal transport , author=. Advances in neural information processing systems , volume=

  71. [71]

    Advances in neural information processing systems , volume=

    Sinkhorn distances: Lightspeed computation of optimal transport , author=. Advances in neural information processing systems , volume=

  72. [72]

    International conference on artificial intelligence and statistics , pages=

    Smooth and sparse optimal transport , author=. International conference on artificial intelligence and statistics , pages=. 2018 , organization=

    2018

  73. [73]

    International conference on machine learning , pages=

    Fast computation of Wasserstein barycenters , author=. International conference on machine learning , pages=. 2014 , organization=

    2014

  74. [74]

    International Conference on Artificial Intelligence and Statistics , pages=

    Learning generative models with sinkhorn divergences , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2018 , organization=

    2018

  75. [75]

    2008 , howpublished =

    Fabian Dornhege , title =. 2008 , howpublished =

    2008

  76. [76]

    Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume=

    A survey on datasets for fairness-aware machine learning , author=. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , volume=. 2022 , publisher=

    2022

  77. [77]

    International Workshop on Advanced Analytics and Learning on Temporal Data , pages=

    Ismail-Fawaz, Ali and Ismail Fawaz, Hassan and Petitjean, Fran. International Workshop on Advanced Analytics and Learning on Temporal Data , pages=. 2023 , organization=

    2023

  78. [78]

    arXiv preprint arXiv:1910.12366 , year=

    Thieves on sesame street! model extraction of bert-based apis , author=. arXiv preprint arXiv:1910.12366 , year=

    work page arXiv 1910

  79. [79]

    Operations research & management science in the age of analytics , pages=

    Wasserstein distributionally robust optimization: Theory and applications in machine learning , author=. Operations research & management science in the age of analytics , pages=. 2019 , publisher=

    2019

  80. [80]

    2018 IEEE international symposium on technologies for homeland security (HST) , pages=

    Active deep learning attacks under strict rate limitations for online API calls , author=. 2018 IEEE international symposium on technologies for homeland security (HST) , pages=. 2018 , organization=

    2018

Showing first 80 references.