Pith. sign in

REVIEW 3 major objections 7 minor 44 references

DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with Limited Data

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read DFF claims that a constrained correction layer can make decision-focused learning safe for limited data and non-differentiable predictors.

desk verdict A genuinely useful fine-tuning wrapper for DFL with non-differentiable backbones, but the 'theoretical guarantee' is the constraint restated and the synthetic gains need error bars. read the letter →

arxiv 2501.01874 v2 pith:TC3FN55D submitted 2025-01-03 cs.LG

Jiaqi Yang , Enming Liang , Zicheng Su , Zhichao Zou , Peng Zhen , Jiecheng Guo , Wanjing Ma , Kun An This is my paper · ORCID
classification cs.LG
keywords decision-focusedlearningpredict-then-optimizefine-tuningtrustregionlimiteddatabiascorrectionmoduleSPO+surrogatelossnon-differentiablemodels
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 argues that decision-focused learning can be made safe for small datasets by fine-tuning the output of an already-trained predictor rather than retraining it. The proposed method, DFF, adds a small input-dependent scaling layer whose corrections are forced to stay within a pointwise percentage $\epsilon$ of the original prediction. Under that constraint, the paper proves the prediction error and the angle of the prediction vector cannot move far from the backbone model, even when data are scarce. The payoff is that non-differentiable or black-box predictors, such as tree ensembles or simulation models, can be improved for downstream optimization without losing their physical meaning.

What carries the argument

The load-bearing object is the bias correction module $F_\theta(x) = \phi(x)\odot M(x) + b(x)$, a hyper-network-style layer that multiplies the backbone's prediction coordinate-wise by a bounded input-dependent scaling and adds a bias (set to zero in the main design). The scaling is produced by an offset sigmoid, $\phi(x) = (1-\epsilon) + 2\epsilon\,\sigma(h(x))$, which guarantees each corrected coordinate lies within $[1-\epsilon, 1+\epsilon]$ of the original coordinate. This explicit trust region is what carries the proof: it turns the fine-tuning constraint into a structural property of the architecture rather than a penalty, and Theorem 1 uses it to bound both the RMSE shift and the angular shift of the predictions. The surrogate gradient from SPO+, $\partial L_{\mathrm{SPO}+}/\partial c = w^*(2\tilde{c}-c) - w^*(c)$, connects this layer to the downstream optimization problem.

What would settle it

On a synthetic linear predict-then-optimize task with a deliberately misspecified backbone and limited data, compute DFF's held-out normalized decision regret and check every corrected prediction against $|\tilde{c}_i - \hat{c}_i| \le \epsilon |\hat{c}_i|$; the central claim stands only if DFF lowers regret while every prediction obeys the bound.

Watch

Extended reading notes

Core claim

DFF's central claim is that a constrained, input-dependent correction layer can convert any backbone predictor into a decision-focused one while keeping predictions close to the original. Concretely, DFF models the corrected prediction as $\tilde{c} = \phi(x) \odot \hat{c}$ with $\phi(x) = (1-\epsilon) + 2\epsilon\,\sigma(h(x))$, so each coordinate satisfies $|\tilde{c}_i - \hat{c}_i| \le \epsilon |\hat{c}_i|$. Training minimizes the SPO+ surrogate of decision regret through the gradient $w^*(2\tilde{c}-c) - w^*(c)$. The paper proves Theorem 1: the RMSE increase is bounded by $\epsilon\sqrt{d}\|\hat{c}\|_2$ and the cosine similarity between $\tilde{c}$ and $\hat{c}$ is at least $\sqrt{1-\epsilon^2}$. In experiments on network flow, portfolio optimization, and a real ride-hailing subsidy allocation problem, DFF reduces normalized decision regret relative to the backbone while keeping MSE nearly unchanged, and it improves the decisions of a non-differentiable simulation model.

Load-bearing premise

The load-bearing assumption is that the SPO+ approximation to decision loss gives a trustworthy direction for updating predictions on a small dataset, and that the original predictor is accurate enough that a small, bounded correction can improve the final decision.

Editorial extensions

If this is right

  • DFF can be applied to any backbone predictor, including non-differentiable tree ensembles and simulation-based models, because it tunes outputs rather than parameters.
  • With 2-fold cross-fitting, DFF attains the lowest normalized decision regret among the compared methods on the network flow and portfolio benchmarks, and it also improves decision regret on the real ride-hailing subsidy allocation problem.
  • The pointwise constraint keeps prediction bias bounded under limited data: the RMSE increase is at most $\epsilon\sqrt{d}\|\hat{c}\|_2$ and the cosine similarity stays above $\sqrt{1-\epsilon^2}$.
  • Predictions from DFF retain the distributional shape of the ground truth, such as the bimodal subsidy conversion rate, whereas a model trained directly on decision loss exhibits a multiplicative shift.
  • A small trust region, for example $\epsilon$ around 0.2, is enough to realize most of the decision improvement while keeping the MSE stable.

Reading between the lines

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

  • An unstated consequence is that DFF's constraint is architecturally enforced by the sigmoid range, so the pointwise bound holds at inference time for any input and not merely on the training set; this makes the method attractive when prediction plausibility is audited.
  • Because the correction is input-dependent, DFF can implicitly learn which samples need larger decision-oriented adjustments and which should stay close to the backbone, which suggests a natural extension where the allowed radius $\epsilon$ is tuned per sample or per decision class.
  • The trust-region formulation invites a comparison with robust optimization: DFF behaves like a worst-case-safe correction around a nominal forecast, and the same machinery could be tested on problems where uncertainty enters the constraints rather than the objective.
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.

Referee Report

3 major / 7 minor

Summary. This paper proposes DFF, a fine-tuning layer that rescales the predictions of an arbitrary backbone model M(x) by an input-dependent factor constrained to [1-epsilon,1+epsilon] per coordinate, and trains the layer with the SPO+ surrogate to minimize decision regret. The authors claim that the constraint strictly bounds prediction bias even with limited data, that the method applies to non-differentiable and black-box models, and that experiments on synthetic network flow/portfolio problems and a real ride-sourcing allocation problem show consistent decision-quality improvements over the backbone and over NN-SPO+.

Significance. The idea of separating a decision-focused correction layer from the backbone is attractive and practically relevant, since it bypasses differentiability requirements and preserves the backbone's predictive distribution to a controllable degree. The pointwise percentage-error constraint is a clean way to implement a trust region, and the cosine-similarity bound in Eq. (14) is a useful auxiliary guarantee. The empirical setting covers meaningful tasks (network flow, portfolio, ride-sourcing allocation) and includes a non-differentiable simulation model. However, the theoretical centerpiece currently contains a dimensionally incorrect bound, and the empirical claims are not supported by any dispersion measure; with variance reported, the method may well be a useful addition to the predict-then-optimize toolbox.

major comments (3)
  1. [Theorem 1 / Eq. (12)] Equation (12) is dimensionally inconsistent and is not the bound that follows from (10). Since |tilde c_i - hat c_i| <= epsilon |hat c_i|, the triangle inequality gives RMSE(tilde c,c) - RMSE(hat c,c) <= RMSE(tilde c,hat c) <= (epsilon / sqrt(d)) ||hat c||_2, not epsilon sqrt(d) ||hat c||_2. The printed right-hand side is a factor d too large and has units of c times d rather than units of c. This is a load-bearing error in the paper's formal contribution and must be corrected; the proof in Appendix A should be rechecked accordingly.
  2. [Case Study / Tables 1 and 2] The text states that all experiments were run 10 times, but no standard deviations, confidence intervals, or significance tests are reported. Several headline differences are very small: in Table 1, DFF improves over the 2-fold Boost backbone by 0.10, 0.07, 0.11, and 0.01 percentage points of NDR across the four synthetic settings, and in Table 2 the real-data improvement over NN-SPO+ is 0.01 percentage point. Without dispersion or paired tests, the central claim that DFF consistently achieves better decision quality is not established; these gaps should be filled before the empirical conclusions can be assessed.
  3. [Constrained Fine-tuning / Eqs. (10)-(11)] The bias bound advertised in the abstract and in Theorem 1 is a direct algebraic consequence of the hard pointwise constraint (10), which is enforced by construction through (11); it holds for every sample and for any amount of data, so it does not provide a statistical, finite-sample, or regret guarantee. The claim that DFF preserves physical meaning even with limited datasets should be reworded as a constraint-satisfaction property or supported by a data-dependent analysis. As it stands, the theorem does not distinguish DFF from any other method that enforces the same pointwise cap, and it does not by itself justify the limited-data motivation.
minor comments (7)
  1. [After Eq. (14)] The typo 'Theorm 1' should be 'Theorem 1'.
  2. [Resource allocation section] The typo 'XGBosst' should be 'XGBoost'.
  3. [Results paragraph, Table 1] The reference 'Table reftab:case1' is unresolved and should be 'Table 1'.
  4. [Sensitivity analysis paragraph] The sentence 'As shown in Table 4' should refer to Figure 4, which contains the sensitivity analysis.
  5. [Loss Function and Training / Eq. (18)] Equation (18) overloads the symbol c: in w*(2tilde c - c) the first argument is the prediction while c also denotes the ground truth; this should be clarified to avoid confusion in the surrogate-gradient derivation.
  6. [Eq. (14)] The cosine lower bound is only meaningful for epsilon <= 1; the text should state this domain explicitly.
  7. [Eq. (10)] The percentage-error constraint divides by hat c_i and is undefined when hat c_i = 0; the paper should state the assumption that backbone predictions are nonzero or define a limiting case.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the DFF bias bound is a direct consequence of the explicit percentage-error constraint, and the decision-improvement claim rests on independent experiments rather than on a self-referential derivation.

full rationale

The paper's derivation chain is self-contained. DFF minimizes decision regret (Eq. 15) over an input-dependent correction layer (Eq. 9), and the trust-region constraint (Eq. 10) is implemented directly by the sigmoid construction in Eq. (11), which enforces the pointwise bound |tilde_c_i - hat_c_i| <= epsilon |hat_c_i|. Theorem 1 then derives RMSE and cosine bounds as algebraic consequences of that construction. This is a legitimate mathematical guarantee about the architecture, not a hidden fit or a prediction that reproduces its own inputs. The central empirical claim, that DFF improves decision quality over the backbone, is tested against independent baselines in Tables 1 and 2, and those comparisons do not reduce to the constraint by construction. The only self-citation, Gao et al. (2024), is used descriptively to reference the DiDi simulation model and is not load-bearing for any formal result. The abstract's phrase 'even with limited datasets' overstates the theorem, since the bound does not depend on sample size, but that is an interpretation issue rather than circularity. Lack of variance or significance reporting is a statistical-robustness concern, not a circularity concern.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central method uses one hand-set hyperparameter, epsilon, and leans on two domain assumptions: the SPO+ surrogate gradient is a valid descent direction for decision loss in the small-data regime, and the backbone model's predictions are reliable enough that the trust region contains a good solution. No invented physical entities are introduced.

free parameters (1)
  • trust-region radius epsilon = 0.5 in synthetic benchmarks; tuned via sensitivity analysis in the real-data experiments
    Epsilon controls the allowed pointwise percentage deviation of the corrected prediction from the backbone prediction. It is a user-set hyperparameter and directly determines the strength of the bias bound in Theorem 1; the paper selects it based on sensitivity analysis rather than an independent validation protocol.
assumptions (3)
  • domain assumption The SPO+ surrogate loss (Eq. 17) provides a valid training signal for the decision loss, with gradient (Eq. 18) used to update theta.
    DFF minimizes L_SPO+ rather than true regret in training. Fisher consistency of SPO+ is established for linear objectives with infinite data, but the paper does not prove surrogate quality inside the trust region or for the portfolio problem's nonlinear constraint.
  • domain assumption The backbone model M is a reliable starting point, so restricting corrected predictions to [1 minus epsilon, 1 plus epsilon] times M(x) preserves physical meaning and can only help decisions.
    The entire trust-region formulation assumes the original predictor is good enough. If the backbone is badly misspecified, the constraint prevents DFF from making the large corrections that could improve decisions.
  • standard math Standard norm and cosine-similarity inequalities used in Theorem 1.
    The proof of Theorem 1 uses standard norm properties, but the printed statement of Eq. (12) appears dimensionally inconsistent without a stated definition of RMSE, and the cosine bound implicitly assumes a sign convention that is not stated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with Limited Data." pith.science (2026). https://pith.science/paper/TC3FN55D

@misc{pith2026250101874,
  author       = {Pith},
  title        = {Pith review of: DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with Limited Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TC3FN55D}},
  note         = {Machine review of arXiv:2501.01874}
}
read the original abstract

Decision-focused learning (DFL) offers an end-to-end approach to the predict-then-optimize (PO) framework by training predictive models directly on decision loss (DL), enhancing decision-making performance within PO contexts. However, the implementation of DFL poses distinct challenges. Primarily, DL can result in deviation from the physical significance of the predictions under limited data. Additionally, some predictive models are non-differentiable or black-box, which cannot be adjusted using gradient-based methods. To tackle the above challenges, we propose a novel framework, Decision-Focused Fine-tuning (DFF), which embeds the DFL module into the PO pipeline via a novel bias correction module. DFF is formulated as a constrained optimization problem that maintains the proximity of the DL-enhanced model to the original predictive model within a defined trust region. We theoretically prove that DFF strictly confines prediction bias within a predetermined upper bound, even with limited datasets, thereby substantially reducing prediction shifts caused by DL under limited data. Furthermore, the bias correction module can be integrated into diverse predictive models, enhancing adaptability to a broad range of PO tasks. Extensive evaluations on synthetic and real-world datasets, including network flow, portfolio optimization, and resource allocation problems with different predictive models, demonstrate that DFF not only improves decision performance but also adheres to fine-tuning constraints, showcasing robust adaptability across various scenarios.

Figures

Figures reproduced from arXiv: 2501.01874 by the authors.

Figure 1
Figure 1. The distribution of predictions generated by differ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the Decision-Focused Fine-tuning Framework. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Sensitivity analysis of parameter ϵ on the decision loss and MSE loss by 24.37% compared to the original simulation model, while the MSE loss remains at a similar level. This again demon￾strates the effectiveness of the constrained fine-tuning de￾sign. Furthermore, we conduct a sensitivity analysis on the parameter ϵ that confines the region in which the predictive model can be adjusted. As shown in [PITH_FULL_IMAG… view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Distribution of predictions with different methods [PITH_FULL_IMAGE:figures/full_fig_p007_3.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 27 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Agrawal, A.; Amos, B.; Barratt, S.; Boyd, S.; Diamond, S.; and Kolter, J. Z. 2019. Differentiable convex optimization layers. Advances in Neural Information Processing Systems, 32

    work page 2019

  4. [4]

    Amos, B.; and Kolter, J. Z. 2017. Optnet: Differentiable optimization as a layer in neural networks. In International Conference on Machine Learning, 136--145

    2017

  5. [5]

    T.; Mukadam, M.; and Amos, B

    Bansal, D.; Chen, R. T.; Mukadam, M.; and Amos, B. 2024. Taskmet: Task-driven metric learning for model learning. Advances in Neural Information Processing Systems, 36

    2024

  6. [6]

    Beichter, M.; Werling, D.; Heidrich, B.; Phipps, K.; Neumann, O.; Friederich, N.; Mikut, R.; and Hagenmeyer, V. 2024. Decision-Focused Retraining of Forecast Models for Optimization Problems in Smart Energy Systems. In Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems, 170--181

    work page 2024

  7. [7]

    Berthet, Q.; Blondel, M.; Teboul, O.; Cuturi, M.; Vert, J.-P.; and Bach, F. 2020. Learning with differentiable pertubed optimizers. Advances in Neural Information Processing Systems, 33: 9508--9519

    work page 2020

  8. [8]

    Bertsimas, D.; and Kallus, N. 2020. From predictive to prescriptive analytics. Management Science, 66(3): 1025--1044

    work page 2020

Show all 44 references
  1. [9]

    Breiman, L. 2001. Random forests. Machine learning, 45: 5--32

    2001

  2. [10]

    Butler, A.; and Kwon, R. H. 2023. Gradient boosting for convex cone predict and optimize problems. Operations Research Letters, 51(1): 79--83

    2023

  3. [11]

    Chernozhukov, V.; Chetverikov, D.; Demirer, M.; Duflo, E.; Hansen, C.; Newey, W.; and Robins, J. 2018. Double/debiased machine learning for treatment and structural parameters. Econometrics Journal, 21: C1--C68

    2018

  4. [12]

    Dalle, G.; Baty, L.; Bouvier, L.; and Parmentier, A. 2022. Learning with combinatorial optimization layers: a probabilistic approach. arXiv preprint arXiv:2207.13513

    2022 arXiv

  5. [13]

    Ding, N.; Qin, Y.; Yang, G.; Wei, F.; Yang, Z.; Su, Y.; Hu, S.; Chen, Y.; Chan, C.-M.; Chen, W.; et al. 2023. Parameter-efficient fine-tuning of large-scale pre-trained language models. Nature Machine Intelligence, 5(3): 220--235

    2023

  6. [14]

    Donti, P.; Amos, B.; and Kolter, J. Z. 2017. Task-based end-to-end model learning in stochastic optimization. Advances in Neural Information Processing Systems, 30

    2017

  7. [15]

    Predict, then Optimize

    Elmachtoub, A. N.; and Grigas, P. 2022. Smart "Predict, then Optimize". Management Science, 68: 9--26

    2022

  8. [16]

    N.; Lam, H.; Zhang, H.; and Zhao, Y

    Elmachtoub, A. N.; Lam, H.; Zhang, H.; and Zhao, Y. 2023. Estimate-then-optimize versus integrated-estimationoptimization: A stochastic dominance perspective. arXiv preprint arXiv:2304.06833

    2023 arXiv

  9. [17]

    N.; Liang, J

    Elmachtoub, A. N.; Liang, J. C. N.; and McNellis, R. 2020. Decision trees for decision-making under the predict-then-optimize framework. In International Conference on Machine Learning, 2858--2867

    2020

  10. [18]

    Fan, Y.; Watkins, O.; Du, Y.; Liu, H.; Ryu, M.; Boutilier, C.; Abbeel, P.; Ghavamzadeh, M.; Lee, K.; and Lee, K. 2024. Reinforcement learning for fine-tuning text-to-image diffusion models. Advances in Neural Information Processing Systems, 36

    2024

  11. [19]

    M.-C.; Lam, W.; Bing, L.; and Collier, N

    Fu, Z.; Yang, H.; So, A. M.-C.; Lam, W.; Bing, L.; and Collier, N. 2023. On the effectiveness of parameter-efficient fine-tuning. Proceedings of the AAAI conference on artificial intelligence, 37(11): 12799--12807

    2023

  12. [20]

    Gao, S.; Ran, Q.; Su, Z.; Wang, L.; Ma, W.; and Hao, R. 2024. Evaluation system for urban traffic intelligence based on travel experiences: A sentiment analysis approach. Transportation Research Part A: Policy and Practice, 187: 104170

    2024

  13. [21]

    Ha, D.; Dai, A.; and Le, Q. V. 2016. Hypernetworks. arXiv preprint arXiv:1609.09106

    2016 arXiv

  14. [22]

    Hu, X.; Lee, J.; and Lee, J. 2024. Two-Stage Predict+ Optimize for MILPs with Unknown Parameters in Constraints. Advances in Neural Information Processing Systems, 36

    2024

  15. [23]

    C.; and Lee, J

    Hu, X.; Lee, J. C.; and Lee, J. H. 2023. Predict+ Optimize for packing and covering LPs with unknown parameters in constraints. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, 3987--3995

    2023

  16. [24]

    Hu, Y.; Kallus, N.; and Mao, X. 2022. Fast rates for contextual linear optimization. Management Science, 68(6): 4236--4245

    2022

  17. [25]

    Kotary, J.; Fioretto, F.; Van Hentenryck, P.; and Wilder, B. 2021. End-to-end constrained optimization learning: A survey. arXiv preprint arXiv:2103.16378

    2021 arXiv

  18. [26]

    Kotary, J.; Fioretto, F.; Van Hentenryck, P.; and Zhu, Z. 2022. End-to-end learning for fair ranking systems. In Proceedings of the ACM Web Conference 2022, 3520--3530

    2022

  19. [27]

    Kurutach, T.; Clavera, I.; Duan, Y.; Tamar, A.; and Abbeel, P. 2018. Model-ensemble trust-region policy optimization. arXiv preprint arXiv:1802.10592

    2018 arXiv

  20. [28]

    Mandi, J.; Bucarey, V.; Tchomba, M. M. K.; and Guns, T. 2022. Decision-focused learning: Through the lens of learning to rank. In International Conference on Machine Learning, 14935--14947

    2022

  21. [29]

    Mandi, J.; and Guns, T. 2020. Interior point solving for lp-based prediction+ optimisation. Advances in Neural Information Processing Systems, 33: 7272--7282

    2020

  22. [30]

    Mandi, J.; Kotary, J.; Berden, S.; Mulamba, M.; Bucarey, V.; Guns, T.; and Fioretto, F. 2023. Decision-focused learning: Foundations, state of the art, benchmark and future opportunities. arXiv preprint arXiv:2307.13565

    2023 arXiv

  23. [31]

    J.; Guns, T.; et al

    Mandi, J.; Stuckey, P. J.; Guns, T.; et al. 2020. Smart predict-and-optimize for hard combinatorial optimization problems. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, 1603--1610

    2020

  24. [32]

    C.; and Cassandras, C

    Queeney, J.; Paschalidis, I. C.; and Cassandras, C. G. 2021. Uncertainty-aware policy optimization: A robust, adaptive trust region approach. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, 9377--9385

    2021

  25. [33]

    Sadana, U.; Chenreddy, A.; Delage, E.; Forel, A.; Frejinger, E.; and Vidal, T. 2024. A survey of contextual optimization methods for decision-making under uncertainty. European Journal of Operational Research

    2024

  26. [34]

    Schulman, J.; Levine, S.; Abbeel, P.; Jordan, M.; and Moritz, P. 2015. Trust region policy optimization. In International Conference on Machine Learning, 1889--1897

    2015

  27. [35]

    Shah, S.; Wang, K.; Wilder, B.; Perrault, A.; and Tambe, M. 2022. Decision-focused learning without decision-making: Learning locally optimized decision losses. Advances in Neural Information Processing Systems, 35: 1320--1332

    2022

  28. [36]

    Shah, S.; Wilder, B.; Perrault, A.; and Tambe, M. 2024. Leaving the nest: Going beyond local loss functions for predict-then-optimize. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 14902--14909

    2024

  29. [37]

    She, Y.; Atzberger, C.; Blake, A.; Gualandi, A.; and Keshav, S. 2024. MAGIC: Modular Auto-encoder for Generalisable Model Inversion with Bias Corrections. arXiv preprint arXiv:2405.18953

    2024

  30. [38]

    Tang, B.; and Khalil, E. B. 2022. Pyepo: A pytorch-based end-to-end predict-then-optimize library for linear and integer programming. arXiv preprint arXiv:2206.14234

    2022 arXiv

  31. [39]

    Wilder, B.; Dilkina, B.; and Tambe, M. 2019. Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, 1658--1665

    2019

  32. [40]

    B.; Liao, S.; and Ba, J

    Wu, Y.; Mansimov, E.; Grosse, R. B.; Liao, S.; and Ba, J. 2017. Scalable trust-region method for deep reinforcement learning using kronecker-factored approximation. Advances in Neural Information Processing Systems, 30

    2017

  33. [41]

    Wu, Z.; Wang, L.; Huang, F.; Zhou, L.; Song, Y.; Ye, C.; Nie, P.; Ren, H.; Hao, J.; He, R.; et al. 2022. A framework for multi-stage bonus allocation in meal delivery platform. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 4195--4203

    2022

  34. [42]

    Zhang, R.; Han, J.; Liu, C.; Gao, P.; Zhou, A.; Hu, X.; Yan, S.; Lu, P.; Li, H.; and Qiao, Y. 2023. Llama-adapter: Efficient fine-tuning of language models with zero-init attention. arXiv preprint arXiv:2303.16199

    2023 arXiv

  35. [43]

    Zhao, K.; Hua, J.; Yan, L.; Zhang, Q.; Xu, H.; and Yang, C. 2019. A unified framework for marketing budget allocation. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 1820--1830

    2019

  36. [44]

    Zharmagambetov, A.; Amos, B.; Ferber, A.; Huang, T.; Dilkina, B.; and Tian, Y. 2024. Landscape surrogate: Learning decision losses for mathematical optimization under partial information. Advances in Neural Information Processing Systems, 36

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.