Hyperparameter Tuning Through Pessimistic Bilevel Optimization

Bo Zeng; Liang Xu; Meltem Apaydin Ustun; Xiaoning Qian

REVIEW 4 major objections 6 minor 25 references

Hyperparameter Tuning Through Pessimistic Bilevel Optimization

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Pessimistic bilevel hyperparameter tuning, which hedges against the worst model among equally optimal training solutions, beats optimistic tuning in test accuracy under limited or shifted data, via a label-flipping relaxation.

desk verdict Original application of pessimistic bilevel optimization to hyperparameter tuning, with a correct 0-1 loss reformulation but a heuristic implementation whose approximation gap is unproven and whose empirical validation is partly circular. read the letter →

arxiv 2412.03666 v1 pith:FUDHGBHH submitted 2024-12-04 cs.LG

Meltem Apaydin Ustun , Liang Xu , Bo Zeng , Xiaoning Qian This is my paper · ORCID

classification cs.LG MSC 90C2668T0590C31

keywords hyperparametertuningpessimisticbileveloptimizationAutoMLsmall-datalearningmodeluncertaintysupportvectormachinesadversarialperturbationtransfer

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 argues that standard optimistic bilevel hyperparameter tuning—which assumes the training algorithm, among equally optimal models, will deliver the one that validates best—fails precisely in the small-data and distribution-shift regimes where tuning matters. It proposes pessimistic bilevel tuning: choose hyperparameters so that even the worst model in the inner-level optimal solution set performs well on validation data. The paper supplies a relaxation that turns this three-level problem into a tractable single-level mixed-integer program for hinge-loss support vector machines, and reports that pessimistic solutions obtain higher test accuracy than optimistic ones on two real-world datasets and in a CNN-feature transfer setting, with the largest gains at the smallest training sizes. If correct, this indicates automated machine learning should hedge against inner-level model uncertainty rather than assume cooperation from the training optimizer.

What carries the argument

The load-bearing mechanism is the tight relaxation of the three-level pessimistic problem due to Zeng (2020): replicate the inner-level variables at the outer level and add the constraint that the inner model's training loss is no worse than the replica's, converting the pessimistic problem into a bilevel min-max problem (Proposition 2). For binary classification with $0$-$1$ loss, Theorem 3 flips validation labels to turn the inner maximization into an empirical-risk minimization, giving a min-min problem. The implementation then uses hinge loss on the subset of validation points that the optimistic hyperplane misclassifies or leaves inside its margin, adds an $\varepsilon$-relaxation of the training-loss constraint to model suboptimal training, and reduces the whole problem to a single-level mixed-integer program via Karush-Kuhn-Tucker conditions and big-M linearization.

What would settle it

On a small synthetic dataset engineered so that the inner SVM training problem has a non-unique optimal face containing a model that misclassifies a validation point outside the optimistic solution's margin, run the proposed algorithm and also compute the true pessimistic optimum by enumerating all optimal face vertices; if the proposed solution's validation loss exceeds the true pessimistic optimum's, the $V_f$ heuristic is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that pessimistic bilevel optimization for hyperparameter tuning is not merely a safer formulation but a computationally achievable and empirically superior one for binary linear classifiers. Theorem 3 shows that under $0$-$1$ loss the pessimistic min-max problem reduces exactly to a min-min empirical-risk problem by flipping validation labels, and the implemented method approximates that with hinge loss on the set of validation points that the optimistic solution misclassifies or leaves inside its margin, together with an $\varepsilon$-relaxation of the training-loss constraint. In the reported experiments, pessimistic tuning achieves higher average test accuracy than optimistic tuning in nearly every small-data configuration—for example, on the breast-cancer dataset with 10 training and 5 validation points, $0.885$ versus $0.553$ accuracy—and keeps its edge when validation data are perturbed like the test data, and when a CNN feature extractor trained on MNIST is transferred to FashionMNIST, where it averages about $0.97$ versus $0.86$.

Load-bearing premise

The method's guarantee rests on the heuristic that flipping the labels of the points the optimistic solution misclassifies or leaves inside its margin produces a stand-in for the true worst-case inner model; the paper does not bound the error of this replacement, so a misspecified flipped set would make the obtained hyperparameters not truly pessimistic.

Editorial extensions

If this is right

With limited training data, pessimistic tuning makes test accuracy largely insensitive to how the scarce data are split into training and validation sets, because it treats the inner model as uncertain rather than trusting a single optimizer's output.
When the validation set resembles the shifted or perturbed test distribution, pessimistic tuning exploits that resemblance better than optimistic tuning, as seen in the perturbed-validation and CNN transfer experiments.
The $\varepsilon$-suboptimality parameter explicitly hedges against approximate inner-level training; by Proposition 5 the outer-level loss increases monotonically with $\varepsilon$, so $\varepsilon$ is a tunable robustness knob for anticipated distribution shift.
Adopting the pessimistic view in AutoML pipelines would replace the default assumption of a cooperative training solver with a worst-case hedge, at the cost of solving a larger mixed-integer program.

Reading between the lines

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

The label-flipping reduction is specific to binary $0$-$1$ classification; extending pessimistic tuning to multiclass or regression would require a different adversarial-validation surrogate, such as worst-case label or loss perturbations.
Because the flipped set is built from the optimistic solution, the procedure's guarantees inherit the optimistic solution's geometry; a natural variant would re-derive the flipped set from a preliminary pessimistic run and compare the resulting test accuracy.
The paper does not give a principled rule for choosing $\varepsilon$, only monotonicity; one could select it by a second validation split, treating $\varepsilon$ as an outer-loop hyperparameter in the same way the paper tunes the box constraint.
Part of the small-data gain may stem from the worst-case view acting as an implicit regularizer; comparing against an ordinary regularized SVM whose box bound is tuned by grid search on the same validation split would isolate the contribution of the bilevel machinery.

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.

Desk Editor's Note

read the letter

Colleague,

The thing to know: this is a genuine first — pessimistic bilevel optimization for hyperparameter tuning — and the theory in Theorem 3 is correct for 0-1 loss. But the method they actually implement (Eq. 24) is a heuristic, and both the theory and the experiments leave a real gap between the stated objective and what is solved.

What's new and good: they formulate the three-level pessimistic problem, then use Zeng's relaxation to convert it to a min-max, then for binary linear classifiers with 0-1 loss they prove an exact equivalence (Theorem 3) that turns the inner max into a min over flipped labels. That's a clever and useful trick. Proposition 5's monotonicity of the ε-suboptimality value is also sound. The experiments cover limited data, perturbed test sets, and a transfer-learning setup with a CNN feature extractor. The results generally show pessimistic solutions keeping a higher floor than optimistic ones in small-data regimes, and the Appendix tables give raw numbers, which is worth credit.

Where it gets soft: the implementation switches to hinge loss and fixes the flip set Vf from the optimistic solution. That breaks the exact equivalence, and there is no bound on the approximation gap. Appendix B.1 tries to validate the heuristic, but it uses the same flipping operation to compute the 'worst-case' value, so the validation is circular. Also, the paper is internally inconsistent about what gets flipped: Section 4.2 says misclassified and marginal points, but Appendix A says they flipped all validation labels for the Cancer dataset in the small-sample experiments. That should have been caught. The empirical claim of consistent superiority is also overstated: in Table A1 several Diabetes rows show optimistic equal or better (e.g., |V|=5, |T|=10, and many rows with equal means), and there are no significance tests, no baselines like standard cross-validation or random search, and the scope is binary linear classifiers only. So the central claim is plausible but not yet proven.

Who is this for: people working on bilevel optimization for AutoML or on robust hyperparameter selection. It deserves a serious referee because the formulation is novel and the 0-1 loss equivalence is a real result, but the paper needs major revision to either prove a bound on the heuristic, remove the circular validation, or temper the empirical claims.

My recommendation: send to peer review with a request for heavy revision, not a desk reject.

Best.

Referee Report

4 major / 6 minor

Summary. The paper proposes a pessimistic bilevel optimization formulation for hyperparameter tuning in binary linear classification, in which the outer level minimizes validation hinge loss against the worst-case inner-level training solution. Following the relaxation of Zeng (2020), the authors convert the three-level problem into a min-max problem, then, for 0-1 loss, into an equivalent min-min problem via label flipping (Theorem 3). For hinge loss, they introduce a heuristic that flips the labels of a set V_f of misclassified and margin points determined from the optimistic solution, and solve the resulting bilevel linear program by a KKT-based single-level MILP. Experiments on two UCI data sets and on a CNN-extracted FashionMNIST transfer task are reported to show that pessimistic solutions outperform optimistic ones with limited training data or perturbed/shifted test data.

Significance. The theoretical reformulation in Theorem 3 for 0-1 loss is clean, and the paper provides a concrete, globally solvable MILP approach for the hinge-loss case. The experimental section is extensive and includes per-run tables, which is valuable for reproducibility. However, the central empirical claim of consistent superiority is contradicted by the paper's own Table A1, and the implemented method in Eq. (24) is a heuristic whose approximation gap to the true pessimistic objective is not quantified. The paper is therefore a promising step but needs substantial revision before its conclusions can be accepted.

major comments (4)

[Section 5.1.1, Table A1] The abstract and Section 5.1.1 state that pessimistic solutions 'consistently' outperform optimistic ones, but Table A1 contains multiple counterexamples for the Diabetes data set. For instance, with |V|=10, |T|=15 the optimistic accuracy is 0.735 versus 0.722 for pessimistic; with |V|=5, |T|=20 it is 0.728 versus 0.720; and with |V|=10, |T|=20 it is 0.739 versus 0.737. These are not within one standard deviation of each other, so the claim of consistent superiority is not supported by the paper's own results. The wording should be moderated or a proper statistical significance analysis should be provided.
[Section 4.2, Eq. (24)] The exact equivalence in Theorem 3 holds only for 0-1 loss, and Corollary 4 relies on a theta-dependent partition (V1, V2, V3). In the implemented method, however, the flip set V_f is computed once from the optimistic solution and then fixed before solving (24). No bound is given for the gap between the true worst-case inner problem in (23) and the fixed-flip approximation (24). Thus the hyperparameters returned by (24) are not established to be pessimistic solutions of (9). Appendix B.1 does not close this gap because the evaluation of the optimistic model under the worst case in (A1) uses 'the previously mentioned flipping operation'—the same heuristic being validated. A direct comparison of (24) against the exact pessimistic objective, or a bound on the approximation error, is required.
[Appendix A vs. Section 4.2] There is a direct inconsistency in the experimental protocol. Appendix A states that for the Cancer data set, 'we have flipped the class labels of all the points in the randomly sampled validation set for each run,' whereas Section 4.2 specifies flipping only the misclassified and marginal points in V_f. This means the reported Cancer results do not correspond to the algorithm described in Section 4.2. The authors should clarify which procedure was actually used and re-run the experiments if necessary, since flipping all validation labels could be the source of the observed robustness gains.
[Table A2] The claim of consistent superiority in perturbed-test experiments is also contradicted by Table A2 for Diabetes with clean validation data: at rho=0.5 the optimistic accuracy is 0.652 versus 0.647 for pessimistic, and at rho=0.6 it is 0.633 versus 0.629. The text in Section 5.1.2 already acknowledges that only epsilon <= 0.4 helps for small perturbations, but the abstract and conclusions still state a general preference for pessimistic solutions. The conclusions should be aligned with the actual performance patterns in the tables.

minor comments (6)

[Section 5.1.2 heading] The heading 'Experimetns with Perturbed Data' contains a typo; it should read 'Experiments with Perturbed Data'.
[Figure 2 caption and text] The caption describes blue and black circles, but the text later mentions 'the blue and gray fitted surfaces.' Please harmonize the color descriptions so the figure is unambiguous.
[Eq. (12)] In Eq. (12), the notation f_{theta*}(x_i) is used inside the outer objective before theta* is defined in the following line. Please reorder the display so theta* is defined before it is used, and use a clearer notation such as f_{theta^*}(x_i).
[Proposition 2] Proposition 2 is stated without proof and is attributed to Zeng (2020). Since this tight relaxation is the foundation of the entire solution approach, the paper should either include a self-contained proof or state the precise conditions from Zeng (2020) under which the relaxation is tight, and verify that the SVM formulation satisfies those conditions.
[Section 4.2, Eq. (23)] In Eq. (23), the replicated variables are denoted with hats (w_hat, b_hat, xi_hat), but in Eq. (24) the outer-level variables are also hatted while the inner-level variables are not; please make the distinction between outer replication and inner decision variables explicit to avoid confusion.
[Terminology] The abbreviation PBL is used in Eq. (3) without an explicit expansion. Please expand it as 'pessimistic bilevel' at first use.

Circularity Check

2 steps flagged · score 4.0 of 10

The implemented pessimistic solver is seeded by the optimistic solution through the flip set Vf, and its solution-quality validation applies the same flipping heuristic to both sides, so the method is not fully self-contained; the held-out test-accuracy claims retain independent content but are not purely pessimistic-bilevel results.

self definitional [Section 4.2, paragraph before Eq. (24)]
"In this implementation, the margin needs to be determined in advance. Hence, we first solve the optimistic bilevel hyperparameter optimization problem, and then compute the margin, 2 /||w||, with respect to the decision hyperplane obtained by solving (21). Then Vf is determined based on the optimistic hyperplane and the corresponding margin."

The implemented 'pessimistic' problem (24) is not the pessimistic problem (22)/(23): its defining set Vf is fixed from the optimistic solution (21) before the inner worst-case model is chosen. Corollary 4 defines V1, V2, V3 with respect to the decision variable theta that is being maximized, but the implementation replaces theta by the already-fitted optimistic hyperplane. Thus the model whose performance is reported as 'pessimistic' is defined in terms of the very optimistic solution it is claimed to outperform. The test-accuracy comparison in Figures 1-6 is therefore between the optimistic solution and an optimistic-seeded adversarial retraining, not between two independent solutions of the same bilevel formulation.
other [Appendix B.1, after Eq. (A1)]
"To solve the problem in (A1), we apply the previously mentioned flipping operation to compute the pessimistic solution and solve the minimization problem over the validation examples with flipped labels."

The only explicit validation of the flipping approximation evaluates the exact worst-case problem (A1) by applying the same flipping heuristic that defines the pessimistic model. The paper claims that the objective value of (A1) 'should be consistently larger than that of the pessimistic one', which is a property of the exact problems; but since both sides are computed with the same approximation, Figure A1 compares two heuristic instances rather than the heuristic against the exact pessimistic value. This cannot establish the approximation gap between (23) and (24), so the solution-quality evidence is self-confirming.

full rationale

The core empirical claim of the paper, that pessimistic bilevel hyperparameter tuning generalizes better than optimistic tuning on limited or shifted data, is not forced by construction: the final single-level problem (25) is different from the optimistic problem (21), and the reported test accuracies are measured on held-out data that were not used to fit either model. However, the paper contains two genuinely self-referential elements that prevent a clean non-circular reading. First, the implemented pessimistic formulation (24) requires the optimistic solution as an input: Vf is the set of misclassified and margin points of the hyperplane obtained from (21). The pessimistic method therefore cannot be defined without first solving the baseline it is supposed to supersede, so the comparison is not between two independent formulations and any observed gains could be an artifact of that particular optimistic-seeded flip set. Second, the paper's only explicit solution-quality check (Appendix B.1) solves the exact worst-case problem (A1) using the same flipping approximation, so Figure A1 is a comparison of the heuristic with itself rather than a validation of the approximation gap; the claimed monotonic relation between the exact problems cannot be verified by an algorithm that already assumes the approximation. The Zeng (2020) relaxation cited for Proposition 2 is load-bearing and is a self-citation by one of the authors, but it is an externally published, parameter-free mathematical scheme with its own stated assumptions, so under the review rules it counts as independent evidence and does not itself raise the score. The lack of any bound connecting (23) to (24) is a correctness and approximation risk rather than circularity, and the possibility remains that the test-accuracy gains are genuine. For these reasons the score is 4: substantial self-referential content in the construction and validation, but the central external claim is not equivalent to its inputs.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on one cited theorem (Zeng 2020), one heuristic approximation (label flipping via optimistic margin), and one domain assumption about non-uniqueness of the inner solution set. The method also introduces one user-chosen parameter epsilon and hand-set box bounds, so the reader pays for these outside the paper's derivation.

free parameters (2)

epsilon (inner-level suboptimality rate) = 0, 0.2, 0.4, 0.6, 0.8 (user-chosen)
Controls the size of the inner-level solution set approximation; the reported superiority of pessimistic solutions over optimistic ones in Figures 4 and 5 holds only for some epsilon values (e.g., epsilon <= 0.4 for Cancer in Figure 4a).
Box bounds on hyperparameters (w_LB, w_UB) = LB=0, UB=1 for UCI; LB=0, UB=2.5 for MNIST/FashionMNIST (Appendix A)
Set by hand, not derived; they define the search space of the outer-level hyperparameters and affect the solutions. No sensitivity analysis is provided.

assumptions (4)

standard math The tight relaxation of the pessimistic bilevel problem from Zeng (2020), used in Proposition 2, is correct and applicable.
The paper cites Zeng (2020) for the equivalence of the three-level pessimistic problem and the bilevel relaxation (10); it does not re-derive the result. This is a peer-reviewed published result, but the paper's whole method rests on it.
standard math LP strong duality and KKT conditions are valid for the inner SVM training problem.
Used to convert the bilevel problem (24) into the single-level MILP (25); valid for linear programs.
domain assumption The inner-level SVM solution set is non-singleton for some hyperparameters in the small-data regime, so the pessimistic and optimistic formulations differ.
Motivates the entire pessimistic approach. The paper infers non-uniqueness from observed performance differences but does not directly verify solution-set cardinality for the datasets.
ad hoc to paper Flipping labels of the optimistic margin/misclassified set Vf yields a valid approximation of the inner worst-case maximization under hinge loss.
This is the key heuristic connecting the exact 0-1-loss reformulation to the implementable hinge-loss MILP (24). No formal approximation bound is given; Appendix B.1 gives only an indirect empirical check.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Hyperparameter Tuning Through Pessimistic Bilevel Optimization." pith.science (2026). https://pith.science/paper/FUDHGBHH

@misc{pith2026241203666,
  author       = {Pith},
  title        = {Pith review of: Hyperparameter Tuning Through Pessimistic Bilevel Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FUDHGBHH}},
  note         = {Machine review of arXiv:2412.03666}
}

read the original abstract

Automated hyperparameter search in machine learning, especially for deep learning models, is typically formulated as a bilevel optimization problem, with hyperparameter values determined by the upper level and the model learning achieved by the lower-level problem. Most of the existing bilevel optimization solutions either assume the uniqueness of the optimal training model given hyperparameters or adopt an optimistic view when the non-uniqueness issue emerges. Potential model uncertainty may arise when training complex models with limited data, especially when the uniqueness assumption is violated. Thus, the suitability of the optimistic view underlying current bilevel hyperparameter optimization solutions is questionable. In this paper, we propose pessimistic bilevel hyperparameter optimization to assure appropriate outer-level hyperparameters to better generalize the inner-level learned models, by explicitly incorporating potential uncertainty of the inner-level solution set. To solve the resulting computationally challenging pessimistic bilevel optimization problem, we develop a novel relaxation-based approximation method. It derives pessimistic solutions with more robust prediction models. In our empirical studies of automated hyperparameter search for binary linear classifiers, pessimistic solutions have demonstrated better prediction performances than optimistic counterparts when we have limited training data or perturbed testing data, showing the necessity of considering pessimistic solutions besides existing optimistic ones.

Figures

Figures reproduced from arXiv: 2412.03666 by the authors.

**Figure 1.** Comparison of pessimistic and optimistic solutions by the average testing accuracy of ten runs with random splits between training and validation sets. Hyperparameter tuning by bilevel optimization can be challenging when we only have limited amount of training data since the learned models can be highly uncertain and prone to overfitting. One critical question in this “small data” regime is: how shall we partition … view at source ↗

**Figure 2.** Comparison of pessimistic and optimistic solutions by the average accuracy with respect to varying total sample size (|V | + |T|) and training to validation splitting ratio (|T|/|V |); Blue circles: performance results of optimistic models, Black circles: performance results of pessimistic models [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗

**Figure 3.** Average testing accuracy for varying training and validation set sizes for the Cancer data set In [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Testing accuracy comparison for various values of ε and perturbation bound ρ on ||∆|| with clean validation data. In [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]

**Figure 5.** Figure 5: Testing accuracy comparison for various values of ε and perturbation bound ρ on ||∆|| with perturbed validation data. MNIST, this rate drops significantly to 0.875 on FashionMNIST when the exactly same model is used. We use the aforementioned CNN model without the last…

**Figure 6.** Figure 6: Comparison of pessimistic and optimistic solutions by the average testing accuracy of five runs with random splits between training and validation sets. 6. Conclusions & Future Research In this paper, we have proposed the first PBL formulation for hyperparameter tuning…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 23 canonical work pages

[1]

Weak linear bilevel programming problems: existence of solutions via a penalty method

Abdelmalek Aboussoror and Abdelatif Mansouri. Weak linear bilevel programming problems: existence of solutions via a penalty method. Journal of Mathematical Analysis and Applications, 304 0 (1): 0 399--408, 2005

work page 2005
[2]

Model selection via bilevel optimization

Kristin P Bennett, Jing Hu, Xiaoyun Ji, Gautam Kunapuli, and Jong-Shi Pang. Model selection via bilevel optimization. In The 2006 IEEE International Joint Conference on Neural Network Proceedings, pages 1922--1929. IEEE, 2006

work page 2006
[3]

A regularization method for ill-posed bilevel optimization problems

Maitine Bergounioux and Mounir Haddou. A regularization method for ill-posed bilevel optimization problems. RAIRO-Operations Research, 40 0 (1): 0 19--35, 2006

work page 2006
[4]

Evasion attacks against machine learning at test time

Battista Biggio, Igino Corona, Davide Maiorca, Blaine Nelson, Nedim S rndi \'c , Pavel Laskov, Giorgio Giacinto, and Fabio Roli. Evasion attacks against machine learning at test time. In Joint European conference on machine learning and knowledge discovery in databases, pages 387--402. Springer, 2013

work page 2013
[5]

Foundations of bilevel programming

Stephan Dempe. Foundations of bilevel programming. Springer Science & Business Media, 2002

2002
[6]

Bilevel optimization: theory, algorithms and applications

Stephan Dempe. Bilevel optimization: theory, algorithms and applications. TU Bergakademie Freiberg, Fakult \"a t f \"u r Mathematik und Informatik, 2018

work page 2018
[7]

Necessary optimality conditions in pessimistic bilevel programming

Stephan Dempe, Boris S Mordukhovich, and Alain B Zemkoho. Necessary optimality conditions in pessimistic bilevel programming. Optimization, 63 0 (4): 0 505--533, 2014

work page 2014
[8]

Generic methods for optimization-based modeling

Justin Domke. Generic methods for optimization-based modeling. In Artificial Intelligence and Statistics, pages 318--326. PMLR, 2012

work page 2012

Show all 25 references

[9]

UCI machine learning repository, 2017

Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive.ics.uci.edu/ml

2017
[10]

Forward and reverse gradient-based hyperparameter optimization

Luca Franceschi, Michele Donini, Paolo Frasconi, and Massimiliano Pontil. Forward and reverse gradient-based hyperparameter optimization. In International Conference on Machine Learning, pages 1165--1173. PMLR, 2017

2017
[11]

Bilevel programming for hyperparameter optimization and meta-learning

Luca Franceschi, Paolo Frasconi, Saverio Salzo, Riccardo Grazzi, and Massimilano Pontil. Bilevel programming for hyperparameter optimization and meta-learning. arXiv preprint arXiv:1806.04910, 2018

2018 arXiv
[12]

Value function based difference-of-convex algorithm for bilevel hyperparameter selection problems

Lucy L Gao, Jane Ye, Haian Yin, Shangzhi Zeng, and Jin Zhang. Value function based difference-of-convex algorithm for bilevel hyperparameter selection problems. In International Conference on Machine Learning, pages 7164--7182. PMLR, 2022

2022
[13]

Deep bilevel learning

Simon Jenni and Paolo Favaro. Deep bilevel learning. In Proceedings of the European conference on computer vision (ECCV), pages 618--633, 2018

2018
[14]

Bilevel model selection for support vector machines

Gautam Kunapuli, K Bennett, Jing Hu, and Jong-Shi Pang. Bilevel model selection for support vector machines. In CRM Proceedings and Lecture Notes, volume 45, pages 129--158, 2008

2008
[15]

Pessimistic bilevel optimization: a survey

June Liu, Yuxin Fan, Zhong Chen, and Yue Zheng. Pessimistic bilevel optimization: a survey. International Journal of Computational Intelligence Systems, 11 0 (1): 0 725--736, 2018

2018
[16]

A generic first-order algorithmic framework for bi-level programming beyond lower-level singleton

Risheng Liu, Pan Mu, Xiaoming Yuan, Shangzhi Zeng, and Jin Zhang. A generic first-order algorithmic framework for bi-level programming beyond lower-level singleton. In International Conference on Machine Learning, pages 6305--6315. PMLR, 2020

2020
[17]

A value-function-based interior-point method for non-convex bi-level optimization

Risheng Liu, Xuan Liu, Xiaoming Yuan, Shangzhi Zeng, and Jin Zhang. A value-function-based interior-point method for non-convex bi-level optimization. In International Conference on Machine Learning, pages 6882--6892. PMLR, 2021

2021
[18]

Weak via strong stackelberg problem: new results

Pierre Loridan and Jacqueline Morgan. Weak via strong stackelberg problem: new results. Journal of global Optimization, 8 0 (3): 0 263--287, 1996

1996
[19]

Gradient-based hyperparameter optimization through reversible learning

Dougal Maclaurin, David Duvenaud, and Ryan Adams. Gradient-based hyperparameter optimization through reversible learning. In International conference on machine learning, pages 2113--2122. PMLR, 2015

2015
[20]

Penalty method for inversion-free deep bilevel optimization

Akshay Mehra and Jihun Hamm. Penalty method for inversion-free deep bilevel optimization. arXiv preprint arXiv:1911.03432, 2019

1911 arXiv
[21]

secml: A python library for secure and explainable machine learning

Marco Melis, Ambra Demontis, Maura Pintor, Angelo Sotgiu, and Battista Biggio. secml: A python library for secure and explainable machine learning. arXiv preprint arXiv:1912.10013, 2019

1912 arXiv
[22]

Hyperparameter optimization with approximate gradient

Fabian Pedregosa. Hyperparameter optimization with approximate gradient. In International conference on machine learning, pages 737--746. PMLR, 2016

2016
[23]

Pessimistic bilevel optimization

Wolfram Wiesemann, Angelos Tsoukalas, Polyxeni-Margarita Kleniati, and Ber c Rustem. Pessimistic bilevel optimization. SIAM Journal on Optimization, 23 0 (1): 0 353--380, 2013

2013
[24]

A practical scheme to compute the pessimistic bilevel optimization problem

Bo Zeng. A practical scheme to compute the pessimistic bilevel optimization problem. INFORMS Journal on Computing, 32 0 (4): 0 1128--1142, 2020

2020
[25]

A solution approach to the weak linear bilevel programming problems

Yue Zheng, Debin Fang, and Zhongping Wan. A solution approach to the weak linear bilevel programming problems. Optimization, 65 0 (7): 0 1437--1449, 2016

2016

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Weak linear bilevel programming problems: existence of solutions via a penalty method

Abdelmalek Aboussoror and Abdelatif Mansouri. Weak linear bilevel programming problems: existence of solutions via a penalty method. Journal of Mathematical Analysis and Applications, 304 0 (1): 0 399--408, 2005

work page 2005

[2] [2]

Model selection via bilevel optimization

Kristin P Bennett, Jing Hu, Xiaoyun Ji, Gautam Kunapuli, and Jong-Shi Pang. Model selection via bilevel optimization. In The 2006 IEEE International Joint Conference on Neural Network Proceedings, pages 1922--1929. IEEE, 2006

work page 2006

[3] [3]

A regularization method for ill-posed bilevel optimization problems

Maitine Bergounioux and Mounir Haddou. A regularization method for ill-posed bilevel optimization problems. RAIRO-Operations Research, 40 0 (1): 0 19--35, 2006

work page 2006

[4] [4]

Evasion attacks against machine learning at test time

Battista Biggio, Igino Corona, Davide Maiorca, Blaine Nelson, Nedim S rndi \'c , Pavel Laskov, Giorgio Giacinto, and Fabio Roli. Evasion attacks against machine learning at test time. In Joint European conference on machine learning and knowledge discovery in databases, pages 387--402. Springer, 2013

work page 2013

[5] [5]

Foundations of bilevel programming

Stephan Dempe. Foundations of bilevel programming. Springer Science & Business Media, 2002

2002

[6] [6]

Bilevel optimization: theory, algorithms and applications

Stephan Dempe. Bilevel optimization: theory, algorithms and applications. TU Bergakademie Freiberg, Fakult \"a t f \"u r Mathematik und Informatik, 2018

work page 2018

[7] [7]

Necessary optimality conditions in pessimistic bilevel programming

Stephan Dempe, Boris S Mordukhovich, and Alain B Zemkoho. Necessary optimality conditions in pessimistic bilevel programming. Optimization, 63 0 (4): 0 505--533, 2014

work page 2014

[8] [8]

Generic methods for optimization-based modeling

Justin Domke. Generic methods for optimization-based modeling. In Artificial Intelligence and Statistics, pages 318--326. PMLR, 2012

work page 2012