Pith. sign in

REVIEW 4 major objections 5 minor 39 references

Decision-Focused Learning in Network Interdiction Games

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Decision-focused learning fails in network interdiction games unless trained on adversarial cost shifts.

desk verdict Clean formal point about DFL failure modes in interdiction games, but the experiments don't support the headline claim because training and test adversaries don't match. read the letter →

arxiv 2608.09036 v1 pith:G5HFYZJO submitted 2026-08-10 cs.GT cs.LGmath.OC

Luca M. Hartmann , Parinaz Naghizadeh This is my paper · ORCID
classification cs.GTcs.LGmath.OC
keywords decision-focusedlearningnetworkinterdictionStackelberggamespredict-then-optimizeshortestpathadversarialtrainingdecisionregretcostestimation
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

This paper claims that decision-focused learning (DFL), the usual best practice for training a prediction model with a downstream optimization in mind, has a structural blind spot in shortest-path network interdiction games. Many cost estimators achieve zero DFL training loss on the plain shortest-path problem, yet their cost orderings flip once the interdictor adds delays to selected arcs, letting the interdictor steer the evader onto worse paths. The paper proves this failure in a two-path toy example and shows empirically that it reverses DFL's usual advantage over prediction-focused learning (PFL). It then proposes Adversarial DFL (A-DFL), which trains on interdicted cost vectors, and gives conditions under which A-DFL's loss is zero only for estimators that also survive interdiction, restoring DFL's advantage on both grid graphs and a real-world transportation network.

What carries the argument

The central object is the decision-equivalence class of the DFL loss. For the toy's affine predictors, the DFL regret vanishes exactly on $\mathcal{F}_{\mathrm{DFL}} = \{ f(w) = a w + b \mid b_1 = b_2,\ a_1 > a_2 \}$, so the training gradient is zero there and the model never learns the true slopes. The A-DFL loss (Eq. 5) averages the DFL regret over shifted costs $c + d \odot x$ for many interdictor choices $x$; on the toy example its zero-loss set is the strictly smaller $\mathcal{F}_{\mathrm{A\text{-}DFL}}$ with $a_1 - a_2 = 2$, so the gradient signal reappears exactly in the region where the nominal loss was flat. The gap between the two classes is the mechanism of the paper: it is the set of estimators that look perfect before interdiction and fail after it.

What would settle it

Run the toy example of Section 5 with $d_1=d_2=3$, train a linear model with DFL to convergence, and evaluate the A-DFL loss (Eq. 5) on the trained model. The paper predicts a nonzero A-DFL loss and a failure interval of feature values where the post-interdiction path choice disagrees with the true optimum; if the trained model satisfies $a_1-a_2=2$, or if no such interval exists, the central failure claim is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms: in a shortest-path network interdiction game, an evader who trains its cost predictor with decision-focused learning can converge to a broad equivalence class of estimators that achieve zero DFL regret on the nominal shortest-path problem yet disagree with the true cost ordering once the interdictor adds delays to chosen arcs. For the two-path toy model with true costs $c(w) = [w, -w]^\top$ and delay vector $d = [3,3]^\top$, the zero-loss class is $\mathcal{F}_{\mathrm{DFL}} = \{ f(w) = a w + b \mid b_1 = b_2,\ a_1 > a_2 \}$; after interdiction, only the strictly smaller class $\mathcal{F}_{\mathrm{A\text{-}DFL}} = \{ f(w) = a w + b \mid b_1 = b_2,\ a_1 - a_2 = 2 \}$ preserves the correct path choice for all $w$. Because the interdictor knows the evader's model, it can pick the interdiction that exploits estimators in $\mathcal{F}_{\mathrm{DFL}} \setminus \mathcal{F}_{\mathrm{A\text{-}DFL}}$, reversing DFL's usual advantage over prediction-focused learning. The paper proposes A-DFL, which replaces the nominal DFL loss with an average of DFL losses over randomly or adversarially sampled interdicted cost vectors, and shows empirically that this collapses the harmful part of the equivalence class and restores DFL's advantage on grid graphs and a real-world transportation network.

Load-bearing premise

The evader must observe the interdictor's chosen arcs and delays before routing, so that the same shifted cost vector $c + d \odot x$ can be fed to the shortest-path solver at both training (A-DFL) and test time; if the evader cannot see interdictions, A-DFL's loss and the whole comparison to DFL become ill-posed.

Editorial extensions

If this is right

  • In any predict-then-optimize pipeline where a game opponent can shift the follower's costs, training on nominal samples alone is not enough; the training distribution must include the opponent's plausible shifts.
  • A-DFL's two variants give a concrete prescription: random interdiction sampling is cheap and effective when the interdiction budget covers a sizable share of arcs, while worst-case (true-cost) selection pays off when the graph is large relative to the budget.
  • The advantage of DFL-based training over PFL in these settings grows with the complexity of the feature-to-cost mapping, so A-DFL's restoration of DFL's edge holds precisely in the regimes where the predictor cannot perfectly fit the truth.
  • The decision-equivalence class analysis gives a diagnostic: if a trained estimator lies in the zero-loss class for nominal decisions but outside the class that survives interdiction, the interdictor can exploit it; A-DFL collapses that gap.

Reading between the lines

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

  • The equivalence-class argument is written for affine predictors; for nonlinear models the failure would likely reappear as flat regions of the DFL loss, and the A-DFL remedy may only partially collapse them if the model can fit shifted costs independently—an extension worth testing.
  • A testable corollary: in games where the evader observes interdictions only partially or stochastically, the A-DFL loss should be replaced by an expectation over the evader's belief about $x$, and the paper's clean zero-loss characterization would degrade gracefully at best.
  • The same competitive dynamic suggests a min-max training scheme for the interdictor: choose interdictions to maximize the evader's regret under the trained model, making the leader's problem a learning-augmented bilevel game; A-DFL is the evader-side counterpart.
  • The size of the zero-loss decision-equivalence class could itself be used as a regularizer—penalizing predictors with large flat regions—offering an alternative to data augmentation with interdicted scenarios.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies decision-focused learning (DFL) in shortest-path network interdiction (SPNI) games, where an evader with imperfect cost information learns a cost predictor and then solves a shortest-path problem under an interdictor's binary fortification actions. The authors argue that DFL's training objective admits a decision-equivalence class of cost estimators that achieve zero nominal regret but make wrong decisions under interdiction, reversing DFL's usual advantage over prediction-focused learning (PFL). They propose Adversarial DFL (A-DFL), which trains on interdicted scenarios, with two variants: RA-DFL (random interdictions) and AA-DFL (worst-case interdictions chosen using true costs). A toy example with affine predictors is analyzed rigorously, and experiments on grid graphs and the Anaheim network are reported to show that A-DFL restores the advantage. The paper also discusses the role of model complexity and data in these findings.

Significance. If the central claim held, the paper would make a useful contribution by identifying a structural failure mode of DFL in game settings and proposing a training-time remedy. The toy analysis in Section 5 and Appendix B is mathematically clean: the definitions of F_DFL and F_A-DFL and the zero-loss characterizations are proven correctly for affine predictors. The paper is also transparent about some limitations, such as the assumption that the evader observes interdictions (Assumption 3(c)) and the restriction of the equivalence-class analysis to linear predictors. The code is publicly available. However, the experimental evidence is too thin and the training/evaluation mismatch is too serious for the general claim to be accepted at face value.

major comments (4)
  1. [4.3 vs 6.2 (train/test interdiction mismatch)] The training distributions of interdictions in A-DFL do not match the test-time adversary. In Section 4.3, RA-DFL samples x uniformly at random and AA-DFL selects x by solving the worst-case problem with the true costs c; neither depends on the trained model. In the evaluation, Section 6.2, the interdictor is the model-aware adversary: x is obtained by solving (1) with the predicted costs ĉ. No theoretical argument, ablation, or diagnostic is provided to show that these distributions coincide or are close in the grid and Anaheim experiments. In the toy example this gap is closed only by the explicit assumption in Section 5.1 that the asymmetric interdictor equals the symmetric interdictor, an assumption that is not verified for the trained DFL model and has no analogue in the larger experiments. Since the central claim is that A-DFL helps the evader against the model-aware interdictor of (1), the reported results do not substantiate that claim without a matched-settings experiment or a bound on the distribution shift.
  2. [Algorithm 1 vs Section 6.1 (confounded data comparison)] The comparison between A-DFL and the PFL/DFL baselines is confounded by the amount of training data. A-DFL (Algorithm 1) is trained on the nominal dataset plus the interdiction dataset, whereas PFL and DFL use only the nominal dataset (Section 6.1). Therefore any observed performance gain of A-DFL over DFL could be an effect of having N_intd additional training samples rather than of the A-DFL loss itself. The paper reports no control experiment, such as training DFL on the same augmented dataset (with the same number of total samples) or training PFL on the interdicted samples. This confound applies to all of Figures 5 and 6 and should be addressed before claiming that A-DFL's advantage comes from its loss.
  3. [Table 1 and Section 6.3 (statistical support)] The experimental claims rely on only N_sim=5 simulations (Table 1), and the paper provides no confidence intervals, standard errors, or significance tests. The text uses the word 'significantly' repeatedly, e.g., in Section 6.3: 'DFL has a significantly wider spread', 'the DFL performance degrades significantly', and 'all DFL methods outperform PFL significantly'. With five seeds, these statements are unsupported. The paper should either report uncertainty estimates (e.g., box plots with per-seed values and confidence intervals) or use a proper statistical test, and qualify any use of 'significant' accordingly.
  4. [Section 6.3, Figure 6(a) vs Abstract/Conclusion] The deg=10 real-world result appears to contradict the paper's headline claim. In Section 6.3 the authors state that for deg=10, 'all DFL methods are outperformed by PFL in the game setting, with the PFL maximum lying more than 5% below the minimum of all other methods', which includes AA-DFL. This is a direct counterexample to the Abstract's claim that A-DFL 'restores DFL's advantage in the game setting without sacrificing performance' and to the summary in Section 7. The paper needs to reconcile this outcome with the central claim, for instance by characterizing the regime in which A-DFL helps, or by softening the claim accordingly.
minor comments (5)
  1. [Section 5.1] The phrase 'without loss of generality' is used for the assumption that the asymmetric and symmetric interdictors coincide; this is a substantive assumption, not a WLOG reduction, and should be phrased as such.
  2. [Equations (8) and (9)] The function-domain notation is inconsistent: Eq. (8) writes f: R^2 -> R^2 while Eq. (9) writes f: R -> R. Both should consistently map from R^2 to R^2.
  3. [Throughout] Typos: 'evador' (Section 1), 'looses' (Abstract), 'mehod' (Section 3.2) should be corrected.
  4. [Section 4.2] The claim that 'our findings are not specific to SPO+ and generalize to other gradient computation methods' is not tested; with no experiments using alternative surrogates, the scope of the claim should be either demonstrated or qualified.
  5. [Table 1] Table 1 reports only 3 random interdiction scenarios per training point (N_intd=3); the sensitivity of RA-DFL to this hyperparameter is not studied, despite the paper's own acknowledgment in Section 7 that N_intd should be systematically studied.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the decision-equivalence class analysis follows from the loss definitions and toy model, and no load-bearing self-citation or fitted-parameter-as-prediction is present.

full rationale

No load-bearing circularity found. Section 5's equivalence classes (Eq. 8: F_DFL = {f(w)=aw+b | b1=b2, a1>a2}; Eq. 9: F_A-DFL = {b1=b2, a1-a2=2}) are derived in Appendix B directly from the DFL regret (Eq. 3), the A-DFL loss (Eq. 5), the toy cost mapping c(w)=[w,-w], the affine model class, and the stated simplified-interdictor assumption. The derivation does not use F_DFL or F_A-DFL as inputs; it characterizes the zero sets of the losses. A-DFL's training objective is the same decision-regret functional used at evaluation (Eq. 5 vs. Section 6.2), but that is the intended definition of decision-focused learning rather than a hidden equivalence, and the paper's comparisons against PFL/DFL baselines plus the toy structural failure provide content beyond the definition. SPO+ is cited to external work (Elmachtoub and Grigas) and is not a self-citation; the only self-reference (Ref. [1], Abdallah/Naghizadeh et al.) is background and not load-bearing. No uniqueness theorem is imported from the authors, and no parameter is fitted and then renamed as a prediction. The skeptic's training/test interdictor mismatch (RA/AA-DFL train on random or true-cost x, while Section 6.2 evaluates against the model-aware interdictor of Eq. 1) is a real external-validity limitation, as are the toy's assumption that asymmetric equals symmetric interdictor and the empirical rather than proven footnote claim that SPO+ training lands in F_DFL; these affect support for the game-setting generalization, but they are not circularity.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central theoretical claim rests on the toy model's affine predictor and the game's informational assumptions. No free parameter is fitted to data beyond the standard training hyperparameters; the listed items are chosen by hand for exposition or set in the experiments. The decision-equivalence class definition itself is derived, not assumed.

free parameters (3)
  • toy delay vector d = [3, 3]
    Chosen by hand in Section 5.1 to fix the delay scale; the resulting F_A-DFL condition is clean, though the relative-slope requirement does not depend on this value.
  • toy model initialization = a1 = 0.1, a2 = -0.1, b = 0
    Chosen in Section 5.1 for visual clarity; Appendix B notes other initializations reach valid members of F_A-DFL.
  • N_intd (random interdiction scenarios) = 3
    Number of random interdiction scenarios per training point in Table 1; sensitivity to this hyperparameter is not studied and is listed as future work.
assumptions (6)
  • domain assumption Assumption 2: the interdictor has full knowledge of the system and can infer the evader's trained model from the same historic data.
    Stated in Section 3.3. Without it, the interdictor's bilevel problem (1) is not well-defined in the learning setting.
  • domain assumption Assumption 3(c): the evader observes the interdictor's actions x and delays d before choosing its route.
    Stated in Section 3.3. A-DFL's decision loss (5) and the test-time evaluation apply the model to the shifted cost \hat{c} + d⊙x; the paper acknowledges in the conclusion that this assumption is restrictive.
  • domain assumption Assumption 4: the interdiction dataset is sampled independently of the nominal dataset and the model's predictions.
    Stated in Section 4.3. It makes gradient computation through the interdiction selection unnecessary.
  • standard math SPO+ surrogate gradient (Eq. 4) from Elmachtoub and Grigas [13] is a valid approximation of the DFL loss gradient.
    Used in all experiments via PyEPO [30]; the paper claims findings are not specific to SPO+.
  • ad hoc to paper In the toy example, the evader's estimates are good enough that the asymmetric interdictor equals the symmetric interdictor.
    Stated in Section 5.1 and detailed in Appendix B. This simplification is only for the toy; it removes the dependence of the interdictor's choice on the estimated costs.
  • domain assumption The data generation process (Eq. 10) with multiplicative noise on [1-ε, 1+ε] models realistic cost-feature relationships.
    Used to generate all synthetic and real-world experiments; the paper notes in the conclusion that real observed cost data remains to be studied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decision-Focused Learning in Network Interdiction Games." pith.science (2026). https://pith.science/paper/G5HFYZJO

@misc{pith2026260809036,
  author       = {Pith},
  title        = {Pith review of: Decision-Focused Learning in Network Interdiction Games},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G5HFYZJO}},
  note         = {Machine review of arXiv:2608.09036}
}
read the original abstract

We study decision-focused learning (DFL) in shortest-path network interdiction (SPNI) games, a Stackelberg game where an interdictor (leader) strengthens the networks' arcs against attacks, while an evader (follower) who is uncertain about costs of attacking network arcs relies on a machine-learned predictor to identify the shortest path. While DFL is highly effective as an end-to-end optimization framework, we show that it faces a fundamental structural failure when employed in this game setting: its training objective admits a broad decision-equivalence class of cost estimators that achieve zero nominal loss yet fail under interdiction, reversing DFL's usual advantage over a naive prediction-focused learning (PFL) approach. To address this, we propose Adversarial DFL (A-DFL), which replaces nominal training samples with interdicted scenarios to collapse the harmful equivalence class. Experiments on synthetic and real-world networks confirm that A-DFL restores DFL's advantage in this game setting, enabling effective end-to-end optimization.

Figures

Figures reproduced from arXiv: 2608.09036 by the authors.

Figure 1
Figure 1. The predict then optimize (PO) decision pipeline. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The toy graph. We choose a linear mapping f : R 2 → R 2 from features w ∈ R 2 to costs c ∈ R 2 :  c1 c2  = f(w) =  w −w  (6) The function f is plotted in Figure 3a. The evader’s problem. We assume the evader uses an affine prediction model of the form cˆ = ˆf(w) = aw + b, where a ∈ R 2 and b ∈ R 2 are learned parameters. The model is trained using either DFL or A-DFL on 5000 samples, 500 epochs. We always initia… view at source ↗
Figure 3
Figure 3. True cost functions (black) and estimated cost functions under DFL (blue) [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Graph topologies used in the experiments. The source node is marked [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Grid Graph Results. This loss can also be interpreted as the percentage increase in cost of the evader’s path over the oracle’s optimal path. Thus, we will use the unit [%] for the simulation regret. When evaluating the game setting, we use new values for the interdict…
Figure 6
Figure 6. Figure 6: Real-World Graph Results. We present A-DFL as a remedy. A-DFL works by augmenting nominal train￾ing samples with interdicted scenarios, and through this, collapses the harmful equivalence class, supplying a gradient signal in exactly the regime where DFL fails. We intr…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 25 canonical work pages

  1. [1]

    IEEE Transactions on Control of Network Systems7(4), 1585–1596 (2020)

    Abdallah,M.,Naghizadeh,P.,Hota,A.R.,Cason,T.,Bagchi,S.,Sundaram,S.:Be- havioral and game-theoretic security investments in interdependent systems mod- eled by attack graphs. IEEE Transactions on Control of Network Systems7(4), 1585–1596 (2020)

    work page 2020

  2. [2]

    Agrawal,A.,Amos,B.,Barratt,S.,Boyd,S.,Diamond,S.,Kolter,Z.:Differentiable convex optimization layers (2019),https://arxiv.org/abs/1910.12430

    arXiv 2019

  3. [3]

    In: Proceedings of the 34th International Conference on Machine Learning

    Amos, B., Kolter, J.Z.: Optnet: Differentiable optimization as a layer in neural net- works. In: Proceedings of the 34th International Conference on Machine Learning. pp.136–145(2017),https://proceedings.mlr.press/v70/amos17a/amos17a.pdf

    work page 2017

  4. [4]

    Computers in Biology and Medicine17(6), 413–422 (1987).https: //doi.org/10.1016/0010-4825(87)90060-6,https://www.sciencedirect.com/ science/article/pii/0010482587900606

    Assimakopoulos, N.: A network interdiction model for hospital infection con- trol. Computers in Biology and Medicine17(6), 413–422 (1987).https: //doi.org/10.1016/0010-4825(87)90060-6,https://www.sciencedirect.com/ science/article/pii/0010482587900606

    work page arXiv 1987

  5. [5]

    Annals of Operations Research335(2), 727–759 (April 2024)

    Azizi, E., Seifi, A.: Shortest path network interdiction with incomplete information: a robust optimization approach. Annals of Operations Research335(2), 727–759 (April 2024)

    work page 2024

  6. [6]

    Networks52(3), 133–140 (2008).https://doi.org/10.1002/net

    Bayrak, H., Bailey, M.D.: Shortest path network interdiction with asymmetric information. Networks52(3), 133–140 (2008).https://doi.org/10.1002/net. 20222

    doi:10.1002/net 2008

  7. [7]

    International Journal of Neural Systems8(4), 433–443 (1997).https://doi.org/ 10.1142/S0129065797000422

    Bengio, Y.: Using a financial training criterion rather than a prediction criterion. International Journal of Neural Systems8(4), 433–443 (1997).https://doi.org/ 10.1142/S0129065797000422

    doi:10.1142/s0129065797000422 1997

  8. [8]

    Management Science66(3), 1025–1044 (2020)

    Bertsimas, D., Kallus, N.: From predictive to prescriptive analytics. Management Science66(3), 1025–1044 (2020)

    work page 2020

Show all 39 references
  1. [9]

    Decision Analysis13(1), 68–98 (2016).https://doi

    Borrero, J.S., Prokopyev, O.A., Sauré, D.: Sequential shortest path interdiction with incomplete information. Decision Analysis13(1), 68–98 (2016).https://doi. org/10.1287/deca.2015.0325,https://doi.org/10.1287/deca.2015.0325

    2016

  2. [10]

    Operations Research57(4), 866–877 (2009).https:// doi.org/10.1287/opre.1080.0643,https://doi.org/10.1287/opre.1080.0643

    Brown,G.G.,Carlyle,W.M.,Harney,R.C.,Skroch,E.M.,Wood,R.K.:Interdicting a nuclear-weapons project. Operations Research57(4), 866–877 (2009).https:// doi.org/10.1287/opre.1080.0643,https://doi.org/10.1287/opre.1080.0643

    2009

  3. [11]

    Complex & Intelligent Systems9(02 2021).https://doi.org/10.1007/ s40747-021-00293-1

    Chu, H., Zhang, W., Bai, P., Chen, Y.: Data-driven optimization for last-mile delivery. Complex & Intelligent Systems9(02 2021).https://doi.org/10.1007/ s40747-021-00293-1

    2021

  4. [12]

    Donti, P.L., Amos, B., Kolter, J.Z.: Task-based end-to-end model learning in stochastic optimization (2019),https://arxiv.org/abs/1703.04529

    2019 arXiv

  5. [13]

    predict, then optimize

    Elmachtoub, A.N., Grigas, P.: Smart "predict, then optimize". Management Sci- ence68(1), 9–26 (2022).https://doi.org/10.1287/mnsc.2020.3922

    2022

  6. [14]

    Networks40(2), 97–111 (2002).https://doi.org/10.1002/net.10039,https://onlinelibrary

    Israeli, E., Wood, R.K.: Shortest-path network interdiction. Networks40(2), 97–111 (2002).https://doi.org/10.1002/net.10039,https://onlinelibrary. wiley.com/doi/abs/10.1002/net.10039

    2002 doi

  7. [15]

    Jenelius, E., Petersen, T., Mattsson, L.G.: Importance and exposure in road net- work vulnerability analysis. Transportation Research Part A: Policy and Practice 40(7), 537–560 (2006).https://doi.org/10.1016/j.tra.2005.11.003,https: //www.sciencedirect.com/science/article/pii/...

    2006 doi

  8. [16]

    In: Proceedings of the 22nd International Conference on Autonomous Agents and MultiAgent Systems, Extended Abstracts (2023) 18 L

    Johnson-Yu, S., Wang, K., Finocchiaro, J., Taneja, A., Tambe, M.: Modeling ro- bustness in decision-focused learning as a stackelberg game. In: Proceedings of the 22nd International Conference on Autonomous Agents and MultiAgent Systems, Extended Abstracts (2023) 18 L. Hartman...

    2023

  9. [17]

    Proceedings of the National Academy of Sciences116(16), 7784–7792 (2019).https://doi.org/10.1073/ pnas.1812459116,https://www.pnas.org/doi/abs/10.1073/pnas.1812459116

    Magliocca, N.R., McSweeney, K., Sesnie, S.E., Tellman, E., Devine, J.A., Nielsen, E.A., Pearson, Z., Wrathall, D.J.: Modeling cocaine traffickers and counterdrug interdiction forces as a complex adaptive system. Proceedings of the National Academy of Sciences116(16), 7784–7792...

    2019 doi

  10. [18]

    Mandi, J., Defresne, M., Berden, S., Guns, T.: Feasibility-aware decision-focused learning for predicting parameters in the constraints (2025),https://arxiv.org/ abs/2510.04951

    2025

  11. [19]

    Journal of Artificial Intelligence Research80, 1623–1701 (Aug 2024).https://doi.org/10.1613/jair.1.15320,http://dx.doi.org/10.1613/ jair.1.15320

    Mandi, J., Kotary, J., Berden, S., Mulamba, M., Bucarey, V., Guns, T., Fioretto, F.: Decision-focused learning: Foundations, state of the art, benchmark and fu- ture opportunities. Journal of Artificial Intelligence Research80, 1623–1701 (Aug 2024).https://doi.org/10.1613/jair...

    2024 doi

  12. [20]

    Mandi, J., Demirović, E., Stuckey, P.J., Guns, T.: Smart predict-and-optimize for hard combinatorial optimization problems (2019),https://arxiv.org/abs/1911. 10092

    2019

  13. [21]

    Computers & Industrial Engineering162, 107708 (2021).https://doi.org/10.1016/j.cie.2021.107708, https://www.sciencedirect.com/science/article/pii/S0360835221006124

    Mirzaei, M., Mirzapour Al-e-hashem, S.M.J., Akbarpour Shirazi, M.: A maximum- flow network interdiction problem in an uncertain environment under information asymmetry condition: Application to smuggling goods. Computers & Industrial Engineering162, 107708 (2021).https://doi.o...

    2021

  14. [22]

    IIE Transactions39(1), 3–14 (2007).https://doi.org/10.1080/07408170500488956, https://doi.org/10.1080/07408170500488956

    Morton, D.P., Pan, F., Saeger, K.J.: Models for nuclear smuggling interdiction. IIE Transactions39(1), 3–14 (2007).https://doi.org/10.1080/07408170500488956, https://doi.org/10.1080/07408170500488956

    2007 doi

  15. [23]

    European Journal of Operational Research297(1), 239–251 (2022).https://doi

    Nguyen, D.H., Smith, J.C.: Network interdiction with asymmetric cost uncertainty. European Journal of Operational Research297(1), 239–251 (2022).https://doi. org/10.1016/j.ejor.2021.04.055,https://www.sciencedirect.com/science/ article/pii/S0377221721003908

    2022 doi

  16. [24]

    In: Proceedings of the 2017 Workshop on Moving Target Defense

    Nguyen, T.H., Wright, M., Wellman, M.P., Baveja, S.: Multi-stage attack graph security games: Heuristic strategies, with empirical game-theoretic analysis. In: Proceedings of the 2017 Workshop on Moving Target Defense. p. 87–97. MTD ’17, Association for Computing Machinery, Ne...

    2017

  17. [25]

    Perrault, A., Wilder, B., Ewing, E., Mate, A., Dilkina, B., Tambe, M.: End-to- end game-focused learning of adversary behavior in security games (2020),https: //arxiv.org/abs/1903.00958

    2020 arXiv

  18. [26]

    Networks83(3), 605–623 (2024).https://doi.org/10.1002/net.22208,https://onlinelibrary

    Punla-Green, S.Z., Mitchell, J.E., Gearhart, J.L., Hart, W.E., Phillips, C.A.: Shortest path network interdiction with asymmetric uncertainty. Networks83(3), 605–623 (2024).https://doi.org/10.1002/net.22208,https://onlinelibrary. wiley.com/doi/abs/10.1002/net.22208

    2024 doi

  19. [27]

    Eu- ropean Journal of Operational Research320(2), 271–289 (2025).https://doi

    Sadana, U., Chenreddy, A., Delage, E., Forel, A., Frejinger, E., Vidal, T.: A survey of contextual optimization methods for decision-making under uncertainty. Eu- ropean Journal of Operational Research320(2), 271–289 (2025).https://doi. org/10.1016/j.ejor.2024.03.020,https://w...

    2025 doi

  20. [28]

    INFORMS Journal on Computing28(3), 527–539 (2016).https://doi.org/10.1287/ijoc.2016.0699, https://doi.org/10.1287/ijoc.2016.0699

    Song, Y., Shen, S.: Risk-averse shortest path interdiction. INFORMS Journal on Computing28(3), 527–539 (2016).https://doi.org/10.1287/ijoc.2016.0699, https://doi.org/10.1287/ijoc.2016.0699

    2016

  21. [29]

    Naval Research Logistics (NRL)61(2), 91–100 Decision-Focused Learning in Network Interdiction Games 19 (2014).https://doi.org/10.1002/nav.21567,https://onlinelibrary.wiley

    Sullivan, K.M., Morton, D.P., Pan, F., Cole Smith, J.: Securing a border un- der asymmetric information. Naval Research Logistics (NRL)61(2), 91–100 Decision-Focused Learning in Network Interdiction Games 19 (2014).https://doi.org/10.1002/nav.21567,https://onlinelibrary.wiley....

    2014 doi

  22. [30]

    Tang, B., Khalil, E.B.: Pyepo: A pytorch-based end-to-end predict-then-optimize library for linear and integer programming (2023),https://arxiv.org/abs/2206. 14234

    2023

  23. [31]

    Socio-Economic Planning Sciences87, 101521 (2023).https://doi

    Tezcan, B., Maass, K.L.: Human trafficking interdiction with decision depen- dent success. Socio-Economic Planning Sciences87, 101521 (2023).https://doi. org/10.1016/j.seps.2023.101521,https://www.sciencedirect.com/science/ article/pii/S0038012123000149

    2023

  24. [32]

    Transportation Networks for Research Core Team: Transportation networks for research.https://github.com/bstabler/TransportationNetworks, accessed Au- gust, 6, 2026

    2026

  25. [33]

    Vlastelica, M., Paulus, A., Musil, V., Martius, G., Rolínek, M.: Differentiation of blackbox combinatorial solvers (2020),https://arxiv.org/abs/1912.02175

    2020 arXiv

  26. [34]

    In: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Sys- tems

    Wang, K., Perrault, A., Mate, A., Tambe, M.: Scalable game-focused learning of adversary models: Data-to-decisions in network security games. In: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Sys- tems. pp. 1449–1457 (2020)

    2020

  27. [35]

    Wang, P.Z., Chen, S., Liang, J., Fioretto, F., Zhu, S.: Gen-dfl: Decision-focused generative learning for robust decision making (2026),https://arxiv.org/abs/ 2502.05468

    2026

  28. [36]

    Wilder, B., Dilkina, B., Tambe, M.: Melding the data-decisions pipeline: Decision- focused learning for combinatorial optimization (2018),https://arxiv.org/abs/ 1809.05504

    2018 arXiv

  29. [37]

    Annals of Operations Research 236(1), 271–289 (January 2016)

    Xu, J., Zhuang, J.: Modeling costly learning and counter-learning in a defender- attacker game with private defender information. Annals of Operations Research 236(1), 271–289 (January 2016)

    2016

  30. [38]

    Transportation Research Part B: Methodological149(C), 52–78 (None 2021)

    Yan, R., Wang, S., Cao, J., Sun, D.: Shipping domain knowledge informed pre- diction and optimization in port state control. Transportation Research Part B: Methodological149(C), 52–78 (None 2021)

    2021

  31. [39]

    Żychowski, A., Mańdziuk, J.: Learning attacker’s bounded rationality model in security games. arXiv preprint arXiv:2109.13036 (2021) A SPO+Safeguard for Graphs with Cycles The SPO+surrogate [13] approximates the decision-loss subgradient asy∗(2ˆc− c)−y ∗(c), which requires sol...

    2021 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.