pith. machine review for the scientific record. sign in

arxiv: 2605.08616 · v1 · submitted 2026-05-09 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

Robust Server Defense Against Unreliable Clients in One-Shot Fair Collaborative Machine Learning

Chia-Yuan Wu , Frank E. Curtis , Daniel P. Robinson
Authors on Pith no claims yet

Pith reviewed 2026-05-12 00:49 UTC · model grok-4.3

classification 💻 cs.LG
keywords collaborative machine learningone-shot learninggroup fairnessunreliable clientsbilevel optimizationserver defenseproxy databiased contributions
0
0 comments X

The pith

A bilevel optimization at the server learns client weights to block biased proxy data and enforce fairness in one-shot collaborative machine learning.

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

The paper addresses one-shot collaborative machine learning, where clients send proxy data only once to train a shared model while preserving privacy. Unreliable clients can supply biased proxy data that favors some groups over others, which harms fairness for minorities without any later chance to correct the bias. The proposed server defense formulates the problem as a bilevel optimization that automatically assigns lower weights to suspect clients and uses a small trusted root dataset held at the server to impose fairness constraints during training. Experiments on standard benchmarks show the approach restores fairness metrics with only minor accuracy cost and continues to work when unreliable clients form the majority. A reader would care because one-shot protocols are increasingly used for efficiency and privacy, yet they leave fairness exposed unless the server can actively defend against bad data.

Core claim

The paper establishes that a bilevel optimization framework learns client-level weights to down-weight biased proxy data contributions while enforcing fairness constraints derived from a small trusted root dataset available at the server, producing a global model whose fairness improves substantially with negligible accuracy loss even when unreliable clients constitute the majority of participants.

What carries the argument

Bilevel optimization that simultaneously optimizes client weights in the outer level and model parameters in the inner level subject to fairness constraints computed on the trusted root dataset.

If this is right

  • The global model maintains group fairness even when unreliable clients are the majority.
  • Accuracy loss remains small compared with an ideal setting that has only reliable clients.
  • The method outperforms prior server-side defenses on the same benchmark tasks.
  • Fairness is enforced without requiring multiple rounds of client-server communication.

Where Pith is reading between the lines

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

  • The framework could be adapted to settings where clients contribute incrementally rather than all at once, provided the server retains the root dataset across rounds.
  • Organizations might combine a small public root dataset with private client proxies to achieve both fairness and privacy in production systems.
  • Performance under real distribution shifts between the root dataset and actual client populations remains an open question that could be tested with domain-specific benchmarks.

Load-bearing premise

A small trusted root dataset exists at the server and accurately represents the desired fairness properties without its own bias or sampling error.

What would settle it

A controlled experiment in which the trusted root dataset is deliberately made biased toward the same groups as the unreliable clients, or in which fairness metrics fail to improve when unreliable clients exceed half the population, would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.08616 by Chia-Yuan Wu, Daniel P. Robinson, Frank E. Curtis.

Figure 1
Figure 1. Figure 1: Illustration of the one-shot CML scenario considered in this paper, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Client-level fairness measures of proxy data for different unreliable [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Bi-criteria comparison of accuracy and fairness across different [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

Collaborative machine learning (CML) enables multiple clients to train a global model jointly in a data-distributed setting. To address data privacy and communication efficiency, one-shot CML has been increasingly adopted, where clients communicate with the server only once by sharing synthetic or processed proxy data. This single-round communication, however, eliminates the possibility of iterative correction at the server, making the learning process particularly vulnerable to client unreliability. In this setting, unreliable clients, whether malicious or non-malicious, may provide biased proxy data that favors certain groups, thereby degrading the fairness of the global model and harming minority or unprivileged groups. In this work, we propose a server-side defense framework based on a bilevel optimization formulation. The proposed approach learns client-level weights to mitigate the influence of biased client proxy data while enforcing fairness constraints by using a very small trusted root dataset available at the server. Experimental results on benchmark datasets show that our method improves fairness with little accuracy loss under biased proxy data contributions from unreliable clients. Moreover, the proposed approach remains effective even when unreliable clients make up a majority of the system, consistently outperforming other existing methods.

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 / 0 minor

Summary. The paper proposes a server-side defense for one-shot collaborative machine learning against unreliable clients that supply biased proxy data. It formulates the problem as a bilevel optimization that learns client-level weights in the outer level while enforcing group fairness constraints in the inner level, using a small trusted root dataset held at the server. The central claims are that the method improves fairness metrics with negligible accuracy degradation on benchmark datasets and remains effective even when unreliable clients constitute a majority, outperforming existing baselines.

Significance. If the experimental claims hold under rigorous validation, the work would be significant for practical deployment of one-shot CML in settings with untrusted or heterogeneous clients, where single-round communication precludes iterative server-side correction. The bilevel formulation that jointly handles weighting and fairness via a trusted root set is a natural technical choice and could generalize to other distributed learning scenarios.

major comments (2)
  1. [Method (bilevel formulation) and Experiments] The load-bearing assumption that the small trusted root dataset is both unbiased and representative of the global test distribution is not subjected to sensitivity analysis or mismatch experiments. When unreliable clients form a majority and supply biased proxies, any demographic mismatch between the root set and the population would allow the learned weights to amplify rather than suppress the very biases the method claims to mitigate. This directly affects the central robustness claim.
  2. [Experiments] The experimental results section provides no quantitative details on root-set size, selection procedure, how bias is injected into client proxies, the specific fairness metrics used, the optimization solver, baseline implementations, or statistical significance testing. Without these, the claims of 'improved fairness with little accuracy loss' and 'consistent outperformance' when unreliable clients are a majority cannot be assessed or reproduced.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our bilevel optimization approach for defending fairness in one-shot collaborative machine learning. The comments highlight important aspects of the trusted root dataset assumption and experimental reproducibility. We address each major comment point by point below.

read point-by-point responses

  1. Referee: [Method (bilevel formulation) and Experiments] The load-bearing assumption that the small trusted root dataset is both unbiased and representative of the global test distribution is not subjected to sensitivity analysis or mismatch experiments. When unreliable clients form a majority and supply biased proxies, any demographic mismatch between the root set and the population would allow the learned weights to amplify rather than suppress the very biases the method claims to mitigate. This directly affects the central robustness claim.

    Authors: We agree that the representativeness of the small trusted root dataset is a central assumption in our bilevel formulation, where it serves as an anchor to enforce group fairness constraints in the inner optimization while learning client weights in the outer level. The manuscript positions this dataset as trusted and held at the server precisely to mitigate the influence of biased proxies from unreliable clients. However, we acknowledge that the current version does not include sensitivity analysis to demographic mismatch between the root set and the global distribution. To strengthen the robustness claims, particularly when unreliable clients are in the majority, we will add new experiments in the revision that systematically vary the root dataset composition (e.g., introducing controlled demographic shifts) and report the resulting fairness metrics and accuracy. This will empirically test whether the learned weights suppress or amplify biases under mismatch. revision: yes

  2. Referee: [Experiments] The experimental results section provides no quantitative details on root-set size, selection procedure, how bias is injected into client proxies, the specific fairness metrics used, the optimization solver, baseline implementations, or statistical significance testing. Without these, the claims of 'improved fairness with little accuracy loss' and 'consistent outperformance' when unreliable clients are a majority cannot be assessed or reproduced.

    Authors: We apologize for the insufficient detail in the experimental section, which is essential for assessing and reproducing the reported improvements in fairness with negligible accuracy loss and outperformance under majority unreliable clients. In the revised manuscript, we will expand this section to provide: root-set sizes (e.g., 100 samples per demographic group on each benchmark), selection procedure (stratified sampling from a trusted pool to maintain balance), bias injection details (specific skew ratios applied to proxy data distributions from unreliable clients), fairness metrics (demographic parity and equalized odds), optimization solver (e.g., Adam optimizer with learning rates, epochs, and convergence criteria for both bilevel levels), baseline implementations (exact adaptations of competing one-shot methods), and statistical testing (means and standard deviations over 5-10 random seeds with significance tests). These additions will directly support the central claims and enable full reproducibility. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation is self-contained

full rationale

The paper presents a bilevel optimization framework that learns client weights while enforcing fairness via a small trusted root dataset at the server. No equations or claims reduce a prediction or result to a fitted parameter by construction, nor does any load-bearing step rely on self-citation chains, imported uniqueness theorems, or ansatzes smuggled from prior author work. Experimental claims rest on benchmark validation rather than tautological redefinitions of inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that a small trusted root dataset exists at the server and can serve as an unbiased anchor for fairness. No explicit free parameters beyond those internal to the bilevel solver are mentioned. No new physical or mathematical entities are introduced.

axioms (2)
  • domain assumption A small trusted root dataset is available at the server and is representative for fairness enforcement
    Invoked to enforce fairness constraints inside the bilevel formulation.
  • domain assumption Bilevel optimization can be solved to produce useful client weights
    Assumed when formulating the defense as a bilevel program.

pith-pipeline@v0.9.0 · 5505 in / 1468 out tokens · 40526 ms · 2026-05-12T00:49:32.008281+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages · 1 internal anchor

  1. [1]

    Collaborative learning: what is it?

    M. Laal and M. Laal, “Collaborative learning: what is it?”Procedia- Social and Behavioral Sciences, vol. 31, pp. 491–495, 2012

    work page 2012

  2. [2]

    Communication-efficient learning of deep networks from decentralized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” inArtificial intelligence and statistics. PMLR, 2017, pp. 1273– 1282

    work page 2017

  3. [3]

    A bilevel optimization approach for computing synthetic data to mitigate unfairness in collab- orative machine learning,

    C.-Y . Wu, F. E. Curtis, and D. P. Robinson, “A bilevel optimization approach for computing synthetic data to mitigate unfairness in collab- orative machine learning,”INFORMS Journal on Optimization, 2026

    work page 2026

  4. [4]

    The current state and challenges of fairness in federated learning,

    S. Vucinich and Q. Zhu, “The current state and challenges of fairness in federated learning,”IEEE Access, vol. 11, pp. 80 903–80 914, 2023

    work page 2023

  5. [5]

    Federated learning for generalization, robustness, fairness: A survey and benchmark,

    W. Huang, M. Ye, Z. Shi, G. Wan, H. Li, B. Du, and Q. Yang, “Federated learning for generalization, robustness, fairness: A survey and benchmark,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 9387–9406, 2024

    work page 2024

  6. [6]

    Fairfed: Enabling group fairness in federated learning,

    Y . H. Ezzeldin, S. Yan, C. He, E. Ferrara, and A. S. Avestimehr, “Fairfed: Enabling group fairness in federated learning,” inProceedings of the AAAI conference on artificial intelligence, vol. 37, 2023, pp. 7494–7502

    work page 2023

  7. [7]

    Fair-fate: Fair federated learning with momentum,

    T. Salazar, M. Fernandes, H. Ara ´ujo, and P. H. Abreu, “Fair-fate: Fair federated learning with momentum,” inInternational Conference on Computational Science. Springer, 2023, pp. 524–538

    work page 2023

  8. [8]

    Fedgan: Federated gen- erative adversarial networks for distributed data,

    M. Rasouli, T. Sun, and R. Rajagopal, “Fedgan: Federated gen- erative adversarial networks for distributed data,”arXiv preprint arXiv:2006.07228, 2020

    work page arXiv 2006

  9. [9]

    One- shot collaborative data distillation,

    W. Holland, C. Thapa, S. A. Siddiqui, W. Shao, and S. Camtepe, “One- shot collaborative data distillation,”arXiv:2408.02266, 2024

    work page arXiv 2024

  10. [10]

    One-shot federated learning,

    N. Guha, A. Talwalkar, and V . Smith, “One-shot federated learning,” arXiv preprint arXiv:1902.11175, 2019

    work page arXiv 1902

  11. [11]

    Dataset distillation.arXiv preprint arXiv:1811.10959, 2018

    T. Wang, J.-Y . Zhu, A. Torralba, and A. A. Efros, “Dataset distillation,” arXiv preprint arXiv:1811.10959, 2018

    work page arXiv 2018

  12. [12]

    Optimized pre-processing for discrimination prevention,

    F. Calmon, D. Wei, B. Vinzamuri, K. Natesan Ramamurthy, and K. R. Varshney, “Optimized pre-processing for discrimination prevention,” Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  13. [13]

    Federated learning with unreliable clients: Performance analysis and mechanism design,

    C. Ma, J. Li, M. Ding, K. Wei, W. Chen, and H. V . Poor, “Federated learning with unreliable clients: Performance analysis and mechanism design,”IEEE Internet of Things Journal, vol. 8, no. 24, pp. 17 308– 17 319, 2021

    work page 2021

  14. [14]

    Vulnerabilities in federated learning,

    N. Bouacida and P. Mohapatra, “Vulnerabilities in federated learning,” IEEe Access, vol. 9, pp. 63 229–63 249, 2021

    work page 2021

  15. [15]

    When federated learning meets blockchain: A new distributed learning paradigm,

    C. Ma, J. Li, L. Shi, M. Ding, T. Wang, Z. Han, and H. V . Poor, “When federated learning meets blockchain: A new distributed learning paradigm,”IEEE Computational Intelligence Magazine, vol. 17, no. 3, pp. 26–33, 2022

    work page 2022

  16. [16]

    Idfl: Incentive-driven federated learning with selfish clients,

    J. Xu, H. Zhang, H. Yu, and G. Fan, “Idfl: Incentive-driven federated learning with selfish clients,”Information Fusion, p. 104185, 2026

    work page 2026

  17. [17]

    Local model poisoning attacks to{Byzantine-Robust}federated learning,

    M. Fang, X. Cao, J. Jia, and N. Gong, “Local model poisoning attacks to{Byzantine-Robust}federated learning,” in29th USENIX security symposium (USENIX Security 20), 2020, pp. 1605–1622

    work page 2020

  18. [18]

    Deep model poisoning attack on federated learning,

    X. Zhou, M. Xu, Y . Wu, and N. Zheng, “Deep model poisoning attack on federated learning,”Future Internet, vol. 13, no. 3, p. 73, 2021

    work page 2021

  19. [19]

    Fldetector: Defending federated learning against model poisoning attacks via detecting ma- licious clients,

    Z. Zhang, X. Cao, J. Jia, and N. Z. Gong, “Fldetector: Defending federated learning against model poisoning attacks via detecting ma- licious clients,” inProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining, 2022, pp. 2545–2555

    work page 2022

  20. [20]

    Data poisoning attacks against federated learning systems,

    V . Tolpegin, S. Truex, M. E. Gursoy, and L. Liu, “Data poisoning attacks against federated learning systems,” inEuropean symposium on research in computer security. Springer, 2020, pp. 480–501

    work page 2020

  21. [21]

    Mitigating data poisoning attacks on a feder- ated learning-edge computing network,

    R. Doku and D. B. Rawat, “Mitigating data poisoning attacks on a feder- ated learning-edge computing network,” inConsumer Communications & Networking Conference (CCNC). IEEE, 2021, pp. 1–6

    work page 2021

  22. [22]

    Illustration icons,

    Flaticon, “Illustration icons,” 2026, available online: https://www.flaticon.com/. Accessed February 4, 2026

    work page 2026

  23. [23]

    Ma- chine learning with adversaries: Byzantine tolerant gradient descent,

    P. Blanchard, E. M. El Mhamdi, R. Guerraoui, and J. Stainer, “Ma- chine learning with adversaries: Byzantine tolerant gradient descent,” Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  24. [24]

    Byzantine-robust dis- tributed learning: Towards optimal statistical rates,

    D. Yin, Y . Chen, R. Kannan, and P. Bartlett, “Byzantine-robust dis- tributed learning: Towards optimal statistical rates,” inInternational conference on machine learning. Pmlr, 2018, pp. 5650–5659

    work page 2018

  25. [25]

    Ro- bust quantity-aware aggregation for federated learning,

    J. Yi, F. Wu, H. Zhang, B. Zhu, T. Qi, G. Sun, and X. Xie, “Ro- bust quantity-aware aggregation for federated learning,”arXiv preprint arXiv:2205.10848, 2022

    work page arXiv 2022

  26. [26]

    Fltrust: Byzantine- robust federated learning via trust bootstrapping,

    X. Cao, M. Fang, J. Liu, and N. Z. Gong, “Fltrust: Byzantine- robust federated learning via trust bootstrapping,”arXiv preprint arXiv:2012.13995, 2020

    work page arXiv 2012

  27. [27]

    Flram: Robust aggregation technique for defense against byzantine poisoning attacks in federated learning,

    H. Chen, X. Chen, L. Peng, and R. Ma, “Flram: Robust aggregation technique for defense against byzantine poisoning attacks in federated learning,”Electronics, vol. 12, no. 21, p. 4463, 2023

    work page 2023

  28. [28]

    Fedcc: Robust federated learning against model poisoning attacks,

    H. Jeong, H. Son, S. Lee, J. Hyun, and T.-M. Chung, “Fedcc: Robust federated learning against model poisoning attacks,”arXiv preprint arXiv:2212.01976, 2022

    work page arXiv 2022

  29. [29]

    Towards addressing label skews in one- shot federated learning,

    Y . Diao, Q. Li, and B. He, “Towards addressing label skews in one- shot federated learning,” inThe Eleventh International Conference on Learning Representations, 2023

    work page 2023

  30. [30]

    One-shot-but-not-degraded federated learning,

    H. Zeng, M. Xu, T. Zhou, X. Wu, J. Kang, Z. Cai, and D. Niyato, “One-shot-but-not-degraded federated learning,” inProceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 11 070– 11 079

    work page 2024

  31. [31]

    Data-free one-shot federated learning under very high statistical heterogeneity,

    C. E. Heinbaugh, E. Luz-Ricca, and H. Shao, “Data-free one-shot federated learning under very high statistical heterogeneity,” inThe Eleventh International Conference on Learning Representations, 2023

    work page 2023

  32. [32]

    Towards one-shot federated learning: Advances, chal- lenges, and future directions,

    F. Amato, L. Qiu, M. Tanveer, S. Cuomo, D. Annunziata, F. Giampaolo, and F. Piccialli, “Towards one-shot federated learning: Advances, chal- lenges, and future directions,”Neurocomputing, p. 132088, 2025

    work page 2025

  33. [33]

    Privacy and fairness in federated learning: On the perspective of tradeoff,

    H. Chen, T. Zhu, T. Zhang, W. Zhou, and P. S. Yu, “Privacy and fairness in federated learning: On the perspective of tradeoff,”ACM Computing Surveys, vol. 56, no. 2, pp. 1–37, 2023

    work page 2023

  34. [34]

    A survey on bias and fairness in machine learning,

    N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman, and A. Galstyan, “A survey on bias and fairness in machine learning,”ACM computing surveys (CSUR), vol. 54, no. 6, pp. 1–35, 2021

    work page 2021

  35. [35]

    A review on fairness in machine learning,

    D. Pessach and E. Shmueli, “A review on fairness in machine learning,” ACM Computing Surveys (CSUR), vol. 55, no. 3, pp. 1–44, 2022

    work page 2022

  36. [36]

    Bias in federated learning: Factors, effects, mitigations, and open issues,

    M. Benmalek and A. Seddiki, “Bias in federated learning: Factors, effects, mitigations, and open issues,”Revue des Sciences et Technolo- gies de l’Information-S´erie ISI: Ing ´enierie des Syst `emes d’Information, vol. 29, no. 6, pp. 2137–2160, 2024

    work page 2024

  37. [37]

    A post-processing-based fair federated learning framework,

    Y . Zhou and N. Goel, “A post-processing-based fair federated learning framework,”arXiv preprint arXiv:2501.15318, 2025

    work page arXiv 2025

  38. [38]

    Glocalfair: Jointly improving global and local group fairness in federated learning,

    S. I. A. Meerza, L. Liu, J. Zhang, and J. Liu, “Glocalfair: Jointly improving global and local group fairness in federated learning,”arXiv preprint arXiv:2401.03562, 2024

    work page arXiv 2024

  39. [39]

    Computationally efficient auto-weighted aggregation for heterogeneous federated learning,

    Z. Talukder and M. A. Islam, “Computationally efficient auto-weighted aggregation for heterogeneous federated learning,” in2022 IEEE inter- national conference on edge computing and communications (EDGE). IEEE, 2022, pp. 12–22

    work page 2022

  40. [40]

    Fed- nolowe: A normalized loss-based weighted aggregation strategy for robust federated learning in heterogeneous environments,

    D.-D. Le, T.-N. Huynh, A.-K. Tran, M.-S. Dao, and P. T. Bao, “Fed- nolowe: A normalized loss-based weighted aggregation strategy for robust federated learning in heterogeneous environments,”PLoS One, vol. 20, no. 8, p. e0322766, 2025

    work page 2025

  41. [41]

    Learning to reweight examples for robust deep learning,

    M. Ren, W. Zeng, B. Yang, and R. Urtasun, “Learning to reweight examples for robust deep learning,” inInternational conference on machine learning. PMLR, 2018, pp. 4334–4343

    work page 2018

  42. [42]

    Fast-convergent federated learning with adaptive weighting,

    H. Wu and P. Wang, “Fast-convergent federated learning with adaptive weighting,”IEEE Transactions on Cognitive Communications and Net- working, vol. 7, no. 4, pp. 1078–1088, 2021

    work page 2021

  43. [43]

    Precision- weighted federated learning,

    J. Reyes, L. Di Jorio, C. Low-Kam, and M. Kersten-Oertel, “Precision- weighted federated learning,”arXiv preprint arXiv:2107.09627, 2021

    work page arXiv 2021

  44. [44]

    Feddwa: Personalized federated learning with dynamic weight adjustment,

    J. Liu, J. Wu, J. Chen, M. Hu, Y . Zhou, and D. Wu, “Feddwa: Personalized federated learning with dynamic weight adjustment,”arXiv preprint arXiv:2305.06124, 2023

    work page arXiv 2023

  45. [45]

    Revisiting weighted aggregation in federated learning with neural networks,

    Z. Li, T. Lin, X. Shang, and C. Wu, “Revisiting weighted aggregation in federated learning with neural networks,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 19 767–19 788

    work page 2023

  46. [46]

    Better safe than sorry: Constructing byzantine-robust federated learning with synthesized trust,

    G. Geng, T. Cai, and Z. Yang, “Better safe than sorry: Constructing byzantine-robust federated learning with synthesized trust,”Electronics, vol. 12, no. 13, p. 2926, 2023

    work page 2023

  47. [47]

    The steepest descent direction for the nonlinear bilevel programming problem,

    G. Savard and J. Gauvin, “The steepest descent direction for the nonlinear bilevel programming problem,”Operations Research Letters, vol. 15, no. 5, pp. 265–272, 1994

    work page 1994

  48. [48]

    A review on bilevel optimization: From classical to evolutionary approaches and applications,

    A. Sinha, P. Malo, and K. Deb, “A review on bilevel optimization: From classical to evolutionary approaches and applications,”IEEE transactions on evolutionary computation, vol. 22, no. 2, pp. 276–295, 2017

    work page 2017

  49. [49]

    Bilevel optimization with a multi-objective lower-level problem: Risk-neutral and risk-averse formulations,

    T. Giovannelli, G. D. Kent, and L. N. Vicente, “Bilevel optimization with a multi-objective lower-level problem: Risk-neutral and risk-averse formulations,”Optimization Methods and Software, pp. 1–23, 2024

    work page 2024

  50. [50]

    Efficient projections onto the l 1-ball for learning in high dimensions,

    J. Duchi, S. Shalev-Shwartz, Y . Singer, and T. Chandra, “Efficient projections onto the l 1-ball for learning in high dimensions,” in Proceedings of the 25th international conference on Machine learning, 2008, pp. 272–279

    work page 2008

  51. [51]

    Fairness in machine learning: A survey,

    S. Caton and C. Haas, “Fairness in machine learning: A survey,”ACM Computing Surveys, vol. 56, no. 7, pp. 1–38, 2024

    work page 2024

  52. [52]

    Fairness constraints: A flexible approach for fair classification,

    M. B. Zafar, I. Valera, M. Gomez-Rodriguez, and K. P. Gummadi, “Fairness constraints: A flexible approach for fair classification,”Journal of Machine Learning Research, vol. 20, no. 75, pp. 1–42, 2019

    work page 2019

  53. [53]

    Lsac national longitudinal bar passage study. lsac research report series

    L. F. Wightman, “Lsac national longitudinal bar passage study. lsac research report series.” ERIC, Tech. Rep., 1998

    work page 1998

  54. [54]

    The 2001 census in the netherlands: Integration of registers and surveys,

    P. Van der Laan, “The 2001 census in the netherlands: Integration of registers and surveys,” inCONFERENCE AT THE CATHIE MARSH CENTRE., 2001, pp. 1–24

    work page 2001

  55. [55]

    Numerical optimization,

    J. Nocedal and S. J. Wright, “Numerical optimization,” 2006

    work page 2006

  56. [56]

    Adam: A Method for Stochastic Optimization

    D. Kingma, “Adam: a method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014