Wasserstein Robust Performative Prediction via Lagrangian Relaxation

Karl H. Johansson; Siyi Wang; Zifan Wang

REVIEW 4 major objections 5 minor 38 references

Wasserstein Robust Performative Prediction via Lagrangian Relaxation

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read The paper proves that two iterative retraining algorithms converge linearly to a unique robust performative stable point when the model is trained against a Wasserstein ambiguity set with model-dependent radius, and bounds how far that poin

desk verdict The main convergence theorems rest on an unproven strong-convexity assumption, so the paper's central claims are not established as written, but the decision-dependent penalty idea is worth revisiting. read the letter →

arxiv 2602.06730 v2 pith:ZH2AIJJP submitted 2026-02-06 math.OC

Siyi Wang , Zifan Wang , Karl H. Johansson This is my paper

classification math.OC MSC 90C2590C1590C47

keywords performativepredictiondistributionallyrobustoptimizationWassersteinambiguitysetLagrangianrelaxationrepeatedriskminimizationgradientdescentstabilitylinearconvergence

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's aim is to show that performative distribution shift—where deploying a model changes the data distribution—can be handled by robust retraining against a Wasserstein ambiguity set. It develops distributionally robust repeated risk minimization and gradient descent, and proves that both converge linearly to a unique robust performative stable point under smoothness, concavity, and sensitivity assumptions; with inexact inner optimization they converge to a bounded neighborhood. The stable point is the fixed point of a Lagrangian-relaxed min-max objective, and the paper bounds its gap to the true performative optimum. If the proof is correct, practitioners can retrain on finite data with explicit convergence rates rather than ad hoc iterations.

What carries the argument

The central mechanism is the Lagrangian relaxation duality (Lemma 1): sup_P (E_{ξ∼P} l(θ,ξ) − λ(θ) W(P, P̂(θ))) = E_{ξ∼P̂(θ)} sup_ζ [l(θ,ζ) − λ(θ)‖ξ−ζ‖²], which converts an infinite-dimensional robust objective into a sample-level min-max problem over adversarial perturbations ζ*(θ,ξ). The proof's contraction arguments then hinge on the robust surrogate φ(θ,ξ,ζ) = l(θ,ζ) − λ(θ)‖ξ−ζ‖² being γ-strongly convex in θ and μ-strongly concave in ζ, together with ε-sensitivity of the distribution map P̂(θ) in Wasserstein-1 distance. The rate κ_rm is exactly the product of that sensitivity and the surrogate's Lipschitz constant, divided by the strong-convexity modulus γ.

What would settle it

Construct a loss that is smooth and concave in ζ but not strongly convex in θ, such as l(θ,ξ)=θ·ξ, with λ(θ)=constant > 0 and compact Ξ; then L_θθ=0 so γ=0 and κ_rm is undefined. If repeated risk minimization still converges linearly in this setting, the theorem's stated Assumptions 1–4 do not suffice; if it fails, the strong-convexity gap is confirmed. A numerical run of Algorithm 1 on this counterexample would settle it.

Watch

Extended reading notes

Core claim

The central claim is that the intractable worst-case distribution objective sup_P E_{ξ∼P}[l(θ,ξ)] with W(P,P̂(θ)) ≤ ρ(θ) can be replaced by the sample-level min-max objective E_{ξ∼P̂(θ)} sup_ζ [l(θ,ζ) − λ(θ)‖ξ−ζ‖²], and that the resulting retraining maps are contractions. Theorem 1 shows ‖G(θ)−G(θ′)‖ ≤ κ_rm‖θ−θ′‖ + C_rm with κ_rm = ε L^f_{θξ}/γ, giving linear convergence to a unique stable point θ_s when κ_rm<1 and to a neighborhood when the inner maximizer is ε-approximate. Theorem 2 extends this to projected gradient descent with κ_gd and a step-size condition. Theorem 3 bounds the myopia gap: ‖θ_s−θ_o‖ ≤ 2εL^f_ξ/γ and the risk gap by 2(εL^f_ξ)²/γ. These are the results the paper is trying

Load-bearing premise

The load-bearing premise is that φ(θ,ξ,ζ) is γ-strongly convex in θ with γ = L_θθ − H_λ D_ξ² > 0; the paper asserts this follows from Assumptions 2 and 4, but those assumptions bound Lipschitz constants from above and do not supply the lower curvature bound that strong convexity requires.

Editorial extensions

If this is right

If the assumptions hold, repeated risk minimization over the empirical distribution converges to a unique robust performative stable point at a linear rate; with inexact inner maximization it converges to a bounded neighborhood.
The same guarantee holds for repeated projected gradient descent when the step size is below the derived threshold, making the method applicable to high-dimensional problems without exact optimization oracles.
The stable point is provably close to the global performative optimum: the parameter gap is O(εL/γ) and the risk gap is O((εL)²/γ), so the performance loss due to myopic retraining is controlled by how sensitive the distribution is to the model.
Allowing the ambiguity radius to depend on θ subsumes constant-radius Wasserstein DRO and gives a principled way to shrink robustness as the model stabilizes.
In the credit-scoring simulation, the distributionally robust model maintains higher defaulter detection than standard performative prediction and a static classifier across strategic sensitivity levels.

Reading between the lines

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

A concrete testable extension is to let the ambiguity radius be learned rather than hand-set; since the theory only requires λ(θ) bounded and smooth with H_λ D_ξ² < L_θθ, any adaptive scheduling scheme that keeps λ inside that envelope should inherit the linear convergence.
The bound ‖θ_s − θ_o‖ ≤ 2εL^f_ξ / γ suggests that the myopia gap is irreducible in the feedback loop, so no retraining algorithm that updates on current data can reach the performative optimum; the robust stable point is the natural attainable target.
If the loss does not have the asserted strong convexity in θ, the contraction denominator γ may vanish; in that regime one would expect either slower, sublinear convergence or oscillation, so the simulations' high-sensitivity spikes are consistent with the theory barely holding.
The min-max surrogate structure connects to adversarial training: ζ*(θ,ξ) can be read as the worst-case feature perturbation, so the framework provides a principled robustness interpretation for strategic feature manipulation in tabular classification.

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

I think the reader's rejection is right: the main theorems don't hold as written. The paper targets a real gap—prior Wasserstein performative prediction work used constant radii and skipped RGD convergence—and the decision-dependent penalty lambda(theta) is a sensible way to shrink ambiguity as the model stabilizes. The credit-scoring simulations are reasonable and suggest the method works empirically. The local Lipschitz computations (Lemma 3, Proposition 1) are mostly standard and appear correct. The load-bearing flaw is the strong-convexity assertion. Section III says phi is gamma-strongly convex in theta with gamma = L_theta_theta - H_lambda D_xi^2, citing Assumptions 2 and 4. Assumption 2 is a Lipschitz upper bound on gradients, not a lower curvature bound; Assumption 3 is concavity in zeta. Nothing in the assumptions gives a positive lower bound on the Hessian of l with respect to theta. The Appendix proof of Theorem 1 even claims strong convexity 'from Assumption 3,' which is a misattribution—concavity in zeta says nothing about convexity in theta. Every contraction inequality and every division by gamma requires gamma > 0, so Theorems 1–3 are unproven. This is not a minor gap; it is the engine of the paper. Two secondary issues reinforce the problem. First, the algorithms sample a single xi_t per iteration, but the convergence analysis treats G as the expected-risk minimizer or expected-gradient step. The theorems do not apply to the actual stochastic iterates. Second, the Lagrangian surrogate with decision-dependent lambda is not shown to be equivalent to the Wasserstein ambiguity-set problem; Lemma 1 covers constant lambda, and the extension needs a proof. The paper does show engagement with the literature and the proof structure suggests the authors know the standard contraction playbook. The flaw may be fixable by adding an explicit strong-convexity assumption and analyzing the stochastic updates separately. But as it stands, the central claims do not follow, and the paper should not be cited for convergence guarantees. Who gets value from this? A researcher working on performative prediction or decision-dependent DRO who wants to repair the theory and push the idea further. It deserves referee time in the sense that a careful referee can identify the missing assumption and guide a revision; I would not accept it in its current form. I would tell the editor: send it to review, but expect substantial revision before any claims can be trusted.

Referee Report

4 major / 5 minor

Summary. The paper considers performative prediction with a Wasserstein distributionally robust objective. It defines a decision-dependent ambiguity radius ρ(θ), relaxes the constrained DRO objective (4) to a Lagrangian-penalized form (5), and uses strong duality (Lemma 1) to obtain a min-max surrogate f(θ,ξ)=sup_ζ l(θ,ζ)-λ(θ)||ξ-ζ||². Building on this surrogate, the authors define a robust performative stable point and propose two algorithms: repeated risk minimization (Algorithm 1) and repeated gradient descent (Algorithm 2). Under Assumptions 1–4, Theorems 1 and 2 claim linear convergence to a unique stable point (or to a neighborhood in the inexact case), and Theorem 3 bounds the suboptimality gap. The paper also presents experiments on a credit-scoring dataset. The main theorems are not supported by the stated assumptions: the required strong convexity of ϕ in θ is asserted but not derived, and the theorems analyze batch expectation updates rather than the single-sample updates in the algorithms.

Significance. If the claims were true, this would be a worthwhile extension of performative prediction to data-driven Wasserstein DRO, particularly the decision-dependent radius and the convergence guarantees for both RRM and RGD. The paper also gives an explicit suboptimality gap and supports the theory with simulations. However, the missing strong-convexity condition is load-bearing in every theorem, and the discrepancy between the stated algorithms and the analyzed updates is substantial. The contribution is therefore not established in the current form.

major comments (4)

[Section III, after Assumption 4; Appendix, Proof of Theorem 1] The paper asserts that ϕ(θ,ξ,ζ) is γ-strongly convex in θ with γ = L_θθ - H_λ D_ξ², citing Assumptions 2 and 4. Assumption 2 is a Lipschitz-smoothness (upper curvature) condition on l; it gives ∇²_θθ l ⪯ L_θθ I, not ⪰. Assumption 3 is concavity in ζ, and Assumption 4 is a Lipschitz/Hessian bound on λ. No lower bound on ∇²_θθ l or on γ is provided. The appendix proof of Theorem 1 explicitly states 'From Assumption 3, ϕ(x,ξ,ζ) is γ-strongly convex in x', which is a misattribution. Since γ appears in the denominators of κ_rm, C_rm, κ_gd, C_gd and in Theorem 3, and the contraction inequalities (27)–(29) require γ>0, the central convergence and suboptimality theorems are not established by the stated assumptions.
[Section III-A, Algorithm 1 and Theorem 1; Section III-B, Algorithm 2 and Theorem 2] The theorems do not analyze the algorithms as written. Algorithm 1 updates θ_{t+1} using a single sample ξ_t via (10), whereas Theorem 1 is stated for the deterministic expectation update (8) and the proof defines G(θ)=argmin J_θ(x) and then takes expectations over ξθ. Similarly, Algorithm 2 uses the single-sample gradient (12)/(13), while Theorem 2's proof analyzes the expected gradient E_{ξθ}[∇J_θ(θ)] in (33)–(40). No stochastic approximation argument (e.g., variance bound, martingale convergence, or high-probability guarantee) is given. Thus the claimed convergence of Algorithm 1 and Algorithm 2 is unproven.
[Section IV, Theorem 3] Theorem 3 depends on the same unsupported strong-convexity property: the proof starts 'Since the function DPR(θ_s,·) is γ-strongly convex'. This is not a consequence of Assumptions 1–4. Also, the θ_o in (7a) is the minimizer of the empirical robust objective DPR(θ,θ), not of the original performative risk (1); the bound (15) therefore does not quantify the gap to the 'global performative optimum' claimed in the abstract. The final theorem is thus not established.
[Section II, Eqs. (4)–(5)] The relation between the constrained DRO problem (4) and the Lagrangian relaxation (5) is not specified. For fixed θ, strong duality relates the constrained problem to min_{λ≥0} [λρ(θ)+E[f]], but (5) fixes λ(θ) a priori as a function of θ. The paper does not state how λ(θ) is chosen from ρ(θ) or prove that minimizing (5) is equivalent to minimizing (4). Consequently the 'robust performative stable point' defined in (7b) is a fixed point of the penalized objective; the connection to the ambiguity-set problem in (4) is not established.

minor comments (5)

[Appendix, proof of Lemma 3, Eq. (17)] The displayed expression for ∇_θϕ(θ,ξ,ζ)-∇_θϕ(θ,ξ',ζ) has a missing parenthesis and incorrectly writes the first term as ∇θl(θ,ζ)-∇θl(θ,ζ).
[Appendix, proof of Theorem 2, Eq. (34)] The term '2η^T T2' should be '2η T2'; T2 is scalar, not a vector.
[Section V, after Eq. (L(θ,ξ))] The symbol γ is used both as the strong-convexity modulus in Section III and as the regularization coefficient in the simulations. This notation clash is confusing; use a different symbol for the regularization parameter.
[Section V, Figs. 2 and 4] The y-axis uses a scaling factor c=||θ^S_0||^{-1}_2 that is not defined in the main text or figure captions. The marker convention for 'numerically zero' also deserves an explicit statement.
[Section II, Assumption 1 and Lemma 2] The ambiguity set uses the Wasserstein distance with squared Euclidean cost, while Assumption 1 and Lemma 2 concern W1. The proof of Theorem 1 uses the W1 Kantorovich-Rubinstein bound, so the metric mismatch should be stated explicitly to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation chain is self-contained and does not reduce to its own inputs; the main limitation is a scope/labeling issue and an unsupported strong-convexity assertion, neither of which is circular.

full rationale

The paper's central claims are the contraction/convergence results (Theorems 1-2) and the suboptimality gap (Theorem 3). These are derived from Assumptions 1-4 plus the cited Lagrangian dual representation (Lemma 1, from Sinha et al. [31]). No equation is reused as its own conclusion: the convergence rates κ_rm=ε L_fθξ/γ and κ_gd depend on the Lipschitz constants computed in Lemma 3 and Proposition 1, and the contraction proofs use strong convexity, smoothness, and ε-sensitivity. The paper contains no self-citations carrying load; all cited tools are external (performative prediction [6], Wasserstein duality [31,32]). The one reduction-like concern is that θ_o and θ_s are defined through the Lagrangian surrogate f and the decoupled risk DPR (Section IV, eq. (14)), so Theorem 3 bounds a gap internal to the relaxed problem (5), not to the original constrained ambiguity-set problem (4). This is a framing/scope limitation, not circularity: the inequality is a nontrivial consequence of γ-strong convexity and ε-sensitivity, not a restatement of the definitions. I also flag a non-circular rigor issue: Section III asserts 'By Assumptions 2 and 4, we establish that φ(θ,ξ,ζ) is γ-strongly convex in θ' with γ=L_θθ-H_λD_ξ^2, and the Appendix proof of Theorem 1 repeats 'From Assumption 3, φ(x,ξ,ζ) is γ-strongly convex in x.' Assumption 2 is an upper smoothness bound (∇²l⪯L_θθ) and Assumption 3 is concavity in ζ; neither supplies the lower curvature in θ needed for γ>0. This affects correctness of Theorems 1-3 but is not a circular dependence between theorem and input. I therefore set circularity score 0.

Assumptions & free parameters 1 free parameters · 7 assumptions · 0 invented entities

The central claim rests on standard DRO duality, the performative-stability fixed-point formulation, and four regularity assumptions. The most important ledger item is the unsupported strong convexity of phi in theta. The penalty schedule lambda(theta) is a hand-chosen free function; the paper never constructs the rho(theta) ambiguity radius it starts with.

free parameters (1)

lambda(theta), decision-dependent penalty schedule = Not fitted; simulations use lambda(theta) = lambda_c + 0.1||theta||^2 with lambda_c in {30, 50, 80} swept
The ambiguity radius rho(theta) in (4) is never instantiated; instead a user-chosen penalty lambda(theta) replaces it. The theory treats lambda_min, lambda_max, L_lambda, H_lambda as given, and the robustness claim changes with this hand-chosen schedule.

assumptions (7)

domain assumption Assumption 1: The distribution map Phat(theta) is epsilon-sensitive in W1 distance.
Controls performative shift and is used in the contraction and suboptimality bounds.
domain assumption Assumption 2: The loss l has Lipschitz continuous gradients in theta and zeta.
Used for smoothness of phi and f in the convergence proofs.
domain assumption Assumption 3: For fixed theta, l(theta, zeta) is concave in zeta.
Used to make phi strongly concave in zeta and the adversarial maximizer unique.
domain assumption Assumption 4: lambda(theta) is bounded, Lipschitz, and has Lipschitz gradient.
Used to bound the effect of the decision-dependent penalty term.
ad hoc to paper Unstated strong convexity of phi(theta, xi, zeta) in theta with modulus gamma = L_theta_theta - H_lambda D_xi^2.
Asserted from Assumptions 2 and 4, but those assumptions do not imply strong convexity. The proof of Theorem 1 cites Assumption 3, which is concavity in zeta. This is load-bearing for Theorems 1-3.
standard math Strong duality / Lemma 1 for the Wasserstein Lagrangian relaxation, from Sinha et al. [31].
Converts the sup over distributions into an expectation of an adversarial surrogate. The equivalence to the constrained ambiguity set with user-specified lambda is not shown.
standard math Kantorovich-Rubinstein duality, Lemma 2, from [32].
Used to bound distribution-shift effects via W1 and Lipschitz functions.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Wasserstein Robust Performative Prediction via Lagrangian Relaxation." pith.science (2026). https://pith.science/paper/ZH2AIJJP

@misc{pith2026260206730,
  author       = {Pith},
  title        = {Pith review of: Wasserstein Robust Performative Prediction via Lagrangian Relaxation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZH2AIJJP}},
  note         = {Machine review of arXiv:2602.06730}
}

read the original abstract

In machine learning, predictive models are trained on historical data. Their deployment may incentivize agents to strategically adapt their behavior, thereby inducing a model-dependent distribution shift. This phenomenon is known as performativity. This paper develops a Wasserstein distributionally robust framework for performative prediction, where the predictive model only has access to limited data. Using these data, we construct an ambiguity set centered on the empirical distribution, and optimize the predictive model against the worst-case distribution. Furthermore, we reformulate the objective as a tractable min-max optimization problem via Lagrangian relaxation, and allow the penalty to depend on the prediction model. Based on this, we develop distributionally robust repeated risk minimization (DR-RRM) and repeated gradient descent (DR-RGD) algorithms to iteratively find a performative stable point amid distributional shifts and model retraining. We theoretically show that both algorithms converge to a stable point linearly under standard regularity conditions. When accounting for approximation errors in the optimization problems, both algorithms converge to a neighborhood of the stable point. Additionally, we establish theoretical bounds on the suboptimality gap between the stable point and the global performative optimum. Finally, numerical simulations of a dynamic credit scoring problem demonstrate the efficacy of the method.

Figures

Figures reproduced from arXiv: 2602.06730 by the authors.

**Figure 1.** illustrates the distributionally robust performative prediction diagram. The decision-maker solves the model parameter θ by minimizing the worst-case risk over a Wasserstein ambiguity set. The deployment of θ then updates the observed data distribution Pˆ(θ) and the risk assessment for the next iteration. Although the primal DRO problem (4) involves an infinitedimensional optimization over probability measures an… view at source ↗

**Figure 2.** Convergence of the distributionally robust performative risk minimization method (Algorithm 1) for varying ε-sensitivity parameters. We add a marker if at the next iteration the distance between iterates is numerically zero. 0.6108 0.6110 0.6112 0.6114 Loss Performance of RRM, =100 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Iteration 0.710 0.712 0.714 0.716 0.718 Accuracy [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Performance evolution of Algorithm 1 under strategic sensitivity ε = 100. Solid blue lines indicate the optimization phase, and dotted green lines indicate the distribution shift after classifier deployment. To account for the discrepancy between the empirical distribution and the true population, we use a distributionally robust framework. Here, the learner minimizes a robust surrogate: min θ∈Θ 1 n Xn i=1 sup ζi∈Ξ… view at source ↗

Figures from the paper (2 more)

**Figure 5.** Figure 5: Top: Transient performance at the first iteration, under performative prediction (PP), distributionally robust performative prediction (DRPP), and a static baseline. ”Pre” and ”post” denote performance before and after the model adapts to the shifted distribution. Bo…

**Figure 6.** Figure 6: Defaulter detection rate under DR-PP and PP after the model adapts to the shifted distribution, with the regulation parameter λ(θ) = λc + 0.1∥θ∥ 2 across 20 iterations. as ε increases. In contrast, both adaptive frameworks (PP and DR-PP) demonstrate the ability to reca…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 2 linked inside Pith

[1]

J. C. Spall,Introduction to Stochastic Search and Optimization: Esti- mation, Simulation, and Control. Hoboken, New Jersey: John Wiley & Sons, 2005

2005
[2]

Letκ rm <1.The sequenceθ t generated by(8)converges linearly to a neighborhood of the unique stable pointθ s with rateκ rm: ∥θt −θ s∥ ≤κt rm∥θ0 −θ s∥+ Crm 1−κ rm , for allt≥0
[3]

The proof of Theorem 1 is provided in the Appendix

When the inner maximization is exact (i.e.,ϵ= 0), the sequenceθ t converges linearly to the fixed pointθ s with rateκ rm. The proof of Theorem 1 is provided in the Appendix. Remark 1.Theorem 1 demonstrates that the iteratesθ t converge faster if the function is smooth and strongly convex (smallerL f θξ, largerγ) and the distribution map is insensitive (sm...
[4]

The sequenceθ t generated by(13)converges linearly to a neighborhood ofθ s with rateκ gd: ∥θt −θ s∥ ≤κt gd∥θ0 −θ s∥+ Cgd 1−κ gd , for allt≥0

Letκ gd <1, i.e., the step sizeη≤ 2(γ−εLf θξ ) β2 . The sequenceθ t generated by(13)converges linearly to a neighborhood ofθ s with rateκ gd: ∥θt −θ s∥ ≤κt gd∥θ0 −θ s∥+ Cgd 1−κ gd , for allt≥0
[5]

Give Me Some Credit

When the inner maximization is exact (i.e.,ϵ= 0), the sequenceθ t converges linearly to the fixed pointθ s with rateκ gd. The proof of Theorem 2 is provided in the Appendix. Remark 2.Similar to Theorem 1, the iteratesθ t converge faster when the objective function is smooth and strongly convex, and when the distribution map exhibits low sensitivity. In th...

2000
[6]

Performa- tive prediction,

J. Perdomo, T. Zrnic, C. Mendler-D ¨unner, and M. Hardt, “Performa- tive prediction,” inInternational Conference on Machine Learning, pp. 7599–7609, 2020

2020
[7]

Zeroth-order optimization meets human feedback: Provable learning via ranking oracles,

Z. Tang, D. Rybin, and T.-H. Chang, “Zeroth-order optimization meets human feedback: Provable learning via ranking oracles,” inInternational Conference on Learning Representations, 2024

2024
[8]

Stochastic optimization of trading strategies in sequential electricity markets,

E. Kraft, M. Russo, D. Keles, and V . Bertsch, “Stochastic optimization of trading strategies in sequential electricity markets,”European Journal of Operational Research, vol. 308, no. 1, pp. 400–421, 2023

2023

Show all 38 references

[9]

Stochastic online opti- mization for cyber-physical and robotic systems,

H. Ma, M. Zeilinger, and M. Muehlebach, “Stochastic online opti- mization for cyber-physical and robotic systems,”Machine Learning, vol. 115, no. 1, p. 11, 2026

2026
[10]

Non-stationary stochastic optimiza- tion,

O. Besbes, Y . Gur, and A. Zeevi, “Non-stationary stochastic optimiza- tion,”Operations Research, vol. 63, no. 5, pp. 1227–1244, 2015. 12 IEEE TRANSACTIONS AND JOURNALS TEMPLATE

2015
[11]

Learning market equilibria preserving statistical privacy using performative prediction,

H. Le Cadre, M. Datar, M. Guckert, and E. Altman, “Learning market equilibria preserving statistical privacy using performative prediction,” IEEE Transactions on Automatic Control, 2025

2025
[12]

Constrained optimization with decision-dependent distributions,

Z. Wang, C. Liu, T. Parisini, M. M. Zavlanos, and K. H. Johansson, “Constrained optimization with decision-dependent distributions,”IEEE Transactions on Automatic Control, 2025

2025
[13]

Online stochas- tic optimization for unknown linear systems: Data-driven controller syn- thesis and analysis,

G. Bianchin, M. Vaquero, J. Cortes, and E. Dall’Anese, “Online stochas- tic optimization for unknown linear systems: Data-driven controller syn- thesis and analysis,”IEEE Transactions on Automatic Control, vol. 69, no. 7, pp. 4411–4426, 2023

2023
[14]

On the stability of iterative retraining of generative models on their own data,

Q. Bertrand, A. J. Bose, A. Duplessis, M. Jiralerspong, and G. Gidel, “On the stability of iterative retraining of generative models on their own data,” inInternational Conference on Learning Representations, 2024

2024
[15]

Multiplayer performative prediction: Learning in decision-dependent games,

A. Narang, E. Faulkner, D. Drusvyatskiy, M. Fazel, and L. J. Ratliff, “Multiplayer performative prediction: Learning in decision-dependent games,”Journal of Machine Learning Research, vol. 24, no. 202, pp. 1– 56, 2023

2023
[16]

Distributionally robust chance constrained data-enabled predictive control,

J. Coulson, J. Lygeros, and F. D ¨orfler, “Distributionally robust chance constrained data-enabled predictive control,”IEEE Transactions on Automatic Control, vol. 67, no. 7, pp. 3289–3304, 2021

2021
[17]

Data-driven distributionally robust optimization using the wasserstein metric: Performance guarantees and tractable reformulations,

P. Mohajerin Esfahani and D. Kuhn, “Data-driven distributionally robust optimization using the wasserstein metric: Performance guarantees and tractable reformulations,”Mathematical Programming, vol. 171, no. 1, pp. 115–166, 2018

2018
[18]

Quantifying distributional model risk via optimal transport,

J. Blanchet and K. Murthy, “Quantifying distributional model risk via optimal transport,”Mathematics of Operations Research, vol. 44, no. 2, pp. 565–600, 2019

2019
[19]

Distributionally robust optimization,

D. Kuhn, S. Shafiee, and W. Wiesemann, “Distributionally robust optimization,”arXiv preprint arXiv:2411.02549, 2024

2024 arXiv
[20]

Wasserstein distributionally robust Kalman filtering,

S. Shafieezadeh-Abadeh, V . A. Nguyen, D. Kuhn, and P. Mohajerin Esfa- hani, “Wasserstein distributionally robust Kalman filtering,” inAdvances in Neural Information Processing Systems, 2018

2018
[21]

Distributionally robustQ-learning,

Z. Liu, Q. Bai, J. Blanchet, P. Dong, W. Xu, Z. Zhou, and Z. Zhou, “Distributionally robustQ-learning,” inInternational Conference on Machine Learning, 2022

2022
[22]

Distributionally robust linear quadratic control,

B. Taskesen, D. Iancu, C ¸ . Koc ¸yi ˘git, and D. Kuhn, “Distributionally robust linear quadratic control,” inAdvances in Neural Information Processing Systems, 2023

2023
[23]

Distributionally robust model predictive control: Closed-loop guarantees and scalable algorithms,

R. D. McAllister and P. M. Esfahani, “Distributionally robust model predictive control: Closed-loop guarantees and scalable algorithms,” IEEE Transactions on Automatic Control, 2024

2024
[24]

Distributionally robust infinite-horizon control: from a pool of samples to the design of dependable controllers,

J.-S. Brouillon, A. Martin, J. Lygeros, F. D ¨orfler, and G. Ferrari- Trecate, “Distributionally robust infinite-horizon control: from a pool of samples to the design of dependable controllers,”IEEE Transactions on Automatic Control, 2025

2025
[25]

However, their framework does not provide convergence guarantees for the resulting iterative pro- cesses

employ Kullback-Leibler (KL) divergence to define the ambiguity set and develop distributionally robust repeated risk minimization (DR-RRM) when access to the explicit form of the function, and repeated gradient descent (DR-RGD) relies on a gradient estimate. However, their fr...
[26]

Wasserstein distributionally robust optimization: Theory and applica- tions in machine learning,

D. Kuhn, P. M. Esfahani, V . A. Nguyen, and S. Shafieezadeh-Abadeh, “Wasserstein distributionally robust optimization: Theory and applica- tions in machine learning,” inOperations Research & Management Science In the Age of Analytics, pp. 130–166, Informs, 2019

2019
[27]

Kullback-leibler divergence constrained distribu- tionally robust optimization,

Z. Hu and L. J. Hong, “Kullback-leibler divergence constrained distribu- tionally robust optimization,”Available at Optimization Online, vol. 1, no. 2, p. 9, 2013

2013
[28]

Statistics of robust op- timization: A generalized empirical likelihood approach,

J. C. Duchi, P. W. Glynn, and H. Namkoong, “Statistics of robust op- timization: A generalized empirical likelihood approach,”Mathematics of Operations Research, vol. 46, no. 3, pp. 946–969, 2021

2021
[29]

Regularization via mass transportation,

S. Shafieezadeh-Abadeh, D. Kuhn, and P. M. Esfahani, “Regularization via mass transportation,”Journal of Machine Learning Research, vol. 20, no. 103, pp. 1–68, 2019

2019
[30]

Distributionally robust performative prediction,

S. Xue and Y . Sun, “Distributionally robust performative prediction,” in Advances in Neural Information Processing Systems, 2024

2024
[31]

Distributionally robust performative optimization,

Z. Jia, Y . Wang, R. Dong, and G. A. Hanasusanto, “Distributionally robust performative optimization,” inAdvances in Neural Information Processing Systems, 2025

2025
[32]

Distributionally robust optimization with decision dependent ambiguity sets,

F. Luo and S. Mehrotra, “Distributionally robust optimization with decision dependent ambiguity sets,”Optimization Letters, vol. 14, no. 8, pp. 2565–2594, 2020

2020
[33]

Decision-dependent distributionally robust optimization with application to dynamic pricing,

C. Qu, H. Jia, and P. You, “Decision-dependent distributionally robust optimization with application to dynamic pricing,” inProc. of the 64th IEEE Conference on Decision and Control, IEEE, 2025

2025
[34]

Distributionally robust opti- mization with decision-dependent ambiguity set,

N. Noyan, G. Rudolf, and M. Lejeune, “Distributionally robust opti- mization with decision-dependent ambiguity set,”Optimization Letters, vol. 14, no. 8, pp. 2541–2564, 2020

2020
[35]

Decision-dependent distributionally robust optimization,

D. Fonseca and M. Junca, “Decision-dependent distributionally robust optimization,”arXiv preprint arXiv:2303.03971, 2023

2023 arXiv
[36]

Certifying some distributional robustness with principled adversarial training,

A. Sinha, H. Namkoong, R. V olpi, and J. C. Duchi, “Certifying some distributional robustness with principled adversarial training,” inInter- national Conference on Learning Representations, 2018

2018
[37]

On the Kantorovich–Rubinstein theorem,

D. A. Edwards, “On the Kantorovich–Rubinstein theorem,”Expositiones Mathematicae, vol. 29, no. 4, pp. 387–398, 2011

2011
[38]

Give me some credit

C. Fusion and W. Cukierski, “Give me some credit.”https:// kaggle.com/competitions/GiveMeSomeCredit, 2011. Kag- gle

2011

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

J. C. Spall,Introduction to Stochastic Search and Optimization: Esti- mation, Simulation, and Control. Hoboken, New Jersey: John Wiley & Sons, 2005

2005

[2] [2]

Letκ rm <1.The sequenceθ t generated by(8)converges linearly to a neighborhood of the unique stable pointθ s with rateκ rm: ∥θt −θ s∥ ≤κt rm∥θ0 −θ s∥+ Crm 1−κ rm , for allt≥0

[3] [3]

The proof of Theorem 1 is provided in the Appendix

When the inner maximization is exact (i.e.,ϵ= 0), the sequenceθ t converges linearly to the fixed pointθ s with rateκ rm. The proof of Theorem 1 is provided in the Appendix. Remark 1.Theorem 1 demonstrates that the iteratesθ t converge faster if the function is smooth and strongly convex (smallerL f θξ, largerγ) and the distribution map is insensitive (sm...

[4] [4]

The sequenceθ t generated by(13)converges linearly to a neighborhood ofθ s with rateκ gd: ∥θt −θ s∥ ≤κt gd∥θ0 −θ s∥+ Cgd 1−κ gd , for allt≥0

Letκ gd <1, i.e., the step sizeη≤ 2(γ−εLf θξ ) β2 . The sequenceθ t generated by(13)converges linearly to a neighborhood ofθ s with rateκ gd: ∥θt −θ s∥ ≤κt gd∥θ0 −θ s∥+ Cgd 1−κ gd , for allt≥0

[5] [5]

Give Me Some Credit

When the inner maximization is exact (i.e.,ϵ= 0), the sequenceθ t converges linearly to the fixed pointθ s with rateκ gd. The proof of Theorem 2 is provided in the Appendix. Remark 2.Similar to Theorem 1, the iteratesθ t converge faster when the objective function is smooth and strongly convex, and when the distribution map exhibits low sensitivity. In th...

2000

[6] [6]

Performa- tive prediction,

J. Perdomo, T. Zrnic, C. Mendler-D ¨unner, and M. Hardt, “Performa- tive prediction,” inInternational Conference on Machine Learning, pp. 7599–7609, 2020

2020

[7] [7]

Zeroth-order optimization meets human feedback: Provable learning via ranking oracles,

Z. Tang, D. Rybin, and T.-H. Chang, “Zeroth-order optimization meets human feedback: Provable learning via ranking oracles,” inInternational Conference on Learning Representations, 2024

2024

[8] [8]

Stochastic optimization of trading strategies in sequential electricity markets,

E. Kraft, M. Russo, D. Keles, and V . Bertsch, “Stochastic optimization of trading strategies in sequential electricity markets,”European Journal of Operational Research, vol. 308, no. 1, pp. 400–421, 2023

2023