pith. sign in

arxiv: 2607.01185 · v1 · pith:RYMSB5DInew · submitted 2026-07-01 · 💻 cs.LG

Neural Certificate Pricing for Combinatorial Optimization Problems

Jingyi Chen , Xinyuan Zhang , Xinwu Qian This is my paper

Pith reviewed 2026-07-02 15:09 UTC · model grok-4.3

classification 💻 cs.LG
keywords neural certificate pricingcombinatorial optimizationdual pricesunsupervised learningamortized separationfeasibility recoveryout-of-distribution generalization
0
0 comments X

The pith

A neural network predicts dual prices for combinatorial optimization such that recovered solutions are globally feasible under a certificate-consistency condition, with first-order price errors inducing only second-order objective loss.

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

The paper introduces Neural Certificate Pricing to handle the exponential search burden in combinatorial optimization problems. A neural network is trained to output certificate-level dual prices while a recovery layer assembles the corresponding primal marginal. The method treats price prediction as amortized separation, learning the aggregate effect of violated inequalities rather than enumerating them. When the certificate-consistency condition holds, the recovered solution is guaranteed to be globally feasible. Local analysis further shows that small errors in the predicted prices affect the objective value only at second order.

Core claim

Neural Certificate Pricing trains a neural network to predict certificate-level dual prices for combinatorial optimization problems. A structured recovery layer then constructs the induced primal marginal. When the certificate-consistency condition holds, the recovered marginal is globally feasible. A local theory establishes that first-order errors in the predicted price induce only second-order loss in objective value.

What carries the argument

Amortized separation via learned residual prices: the neural network predicts certificate-level dual prices whose aggregate effect enters the recovery layer to construct the primal marginal.

If this is right

  • The recovered marginal is globally feasible whenever the certificate-consistency condition holds.
  • First-order errors in the predicted prices produce only second-order losses in objective value.
  • NCP either outperforms state-of-the-art neural baselines by large margins or matches their performance at a fraction of the computation time.
  • NCP exhibits stronger out-of-distribution generalization than the baselines across the three problem classes tested.

Where Pith is reading between the lines

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

  • The same price-prediction approach could be tested on combinatorial problems whose separation oracles are expensive but still polynomial-time verifiable.
  • Integration with existing branch-and-cut frameworks might allow the learned prices to guide branching or cutting decisions without full re-optimization.
  • Empirical checks on problem sizes beyond those in the paper would clarify how far the local second-order error theory extends in practice.

Load-bearing premise

The certificate-consistency condition holds for the neural network's predicted prices.

What would settle it

An instance where the neural network's price predictions violate certificate consistency and the recovered marginal is infeasible, or where first-order price perturbations produce first-order rather than second-order degradation in objective value.

Figures

Figures reproduced from arXiv: 2607.01185 by Jingyi Chen, Xinwu Qian, Xinyuan Zhang.

Figure 1
Figure 1. Figure 1: Left: Schematic illustration of NCP. NCP optimizes the problem inside MG and uses perturbation Γ (price signal) to move the marginal µ to µ ∗ while being consistent to the structure certificate S(y). Right: Maximum-independent set instance. NCP perturbs local node values, then recovers a certificate-consistent marginal from which an independent set can be recovered. through structured search over the expon… view at source ↗
Figure 2
Figure 2. Figure 2: Structural diagnostics on GAP. Left: test-set verification of Theorem 4.2. Around the trained Γ¯ = Γθ(G), additive perturbations of magnitude ε induce O(ε) drift in the certificate state λ and O(ε 2 ) drift in the reduced objective Φ, matching the predicted slopes of 1 and 2 over four orders of magnitude in ε. Middle: non-zero marginal CDFs on instance c1060-1: the LP-relaxation marginal µLP is spread acro… view at source ↗
read the original abstract

Combinatorial optimization (CO) problems are difficult because certifiable discrete structure induces exponential search. One needs to search over the set exponentially many candidates to certify optimality, however, the structural feasibility of a path, packing, or cover can be verified in polynomial time once supplied. In this study, we introduce Neural Certificate Pricing (NCP) that exploits this asymmetry under an unsupervised learning framework. A neural network is trained to predict certificate-level dual prices, while a structured recovery layer constructs the induced primal marginal. NCP can be viewed as amortized separation: instead of enumerating violated inequalities, it learns the residual prices through which their aggregate effect enters recovery. When the certificate-consistency condition holds, the recovered marginal is globally feasible, and a local theory shows that first-order errors in the predicted price induce only second-order loss in objective value. Across three classes of CO problems, NCP either outperforms state-of-the-art neural baselines by large margins or matches them at a fraction of the computation time, and shows stronger out-of-distribution generalization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces Neural Certificate Pricing (NCP) for combinatorial optimization. A neural network is trained unsupervised to predict certificate-level dual prices; a structured recovery layer then constructs the induced primal marginal. The central claims are that, when a certificate-consistency condition holds, the recovered marginal is globally feasible and a local theory shows first-order price errors induce only second-order objective loss. Experiments across three CO problem classes report that NCP either outperforms neural baselines by large margins or matches them at lower compute, with stronger out-of-distribution generalization.

Significance. If the consistency condition can be shown to hold for the learned prices, the framework supplies a principled amortized separation method together with an explicit local error bound; this would be a substantive advance over existing neural CO approaches that lack such feasibility or error guarantees. The unsupervised training and recovery-layer design are potentially reusable strengths.

major comments (3)
  1. [Abstract] Abstract: the claim that 'when the certificate-consistency condition holds, the recovered marginal is globally feasible' and that 'a local theory shows that first-order errors ... induce only second-order loss' is presented without any derivation, proof sketch, or pointer to the section containing the argument. Both the feasibility guarantee and the error result are conditional on this premise, making its substantiation load-bearing.
  2. [NCP framework] NCP framework section: the manuscript trains the network to output prices and then applies recovery, yet supplies no architectural constraint, auxiliary loss term, or post-hoc projection that enforces the certificate-consistency condition. Without such enforcement or a post-training verification step, it is unclear whether the condition holds on instances seen at test time.
  3. [Experimental results] Experimental results section: performance tables and generalization plots are reported, but no ablation or diagnostic is included that checks whether the recovered marginals satisfy the feasibility claim or whether the consistency condition is satisfied by the trained prices on the evaluated instances.
minor comments (2)
  1. [Notation / Framework] Add an explicit equation or boxed definition for the certificate-consistency condition at the first point it is invoked, rather than leaving it as an informal hypothesis.
  2. [Figures] Figure captions for the recovery-layer diagram should state the precise input-output mapping and any assumptions required for the layer to produce a feasible marginal.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight opportunities to strengthen the presentation of the theoretical guarantees and to supply empirical verification of the consistency condition. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'when the certificate-consistency condition holds, the recovered marginal is globally feasible' and that 'a local theory shows that first-order errors ... induce only second-order loss' is presented without any derivation, proof sketch, or pointer to the section containing the argument. Both the feasibility guarantee and the error result are conditional on this premise, making its substantiation load-bearing.

    Authors: The abstract is a high-level summary. The derivation of global feasibility (Theorem 1) and the local second-order error bound (Theorem 2) appears in Section 3. We will revise the abstract to include an explicit pointer to Section 3. revision: yes

  2. Referee: [NCP framework] NCP framework section: the manuscript trains the network to output prices and then applies recovery, yet supplies no architectural constraint, auxiliary loss term, or post-hoc projection that enforces the certificate-consistency condition. Without such enforcement or a post-training verification step, it is unclear whether the condition holds on instances seen at test time.

    Authors: The certificate-consistency condition is a sufficient condition for the stated guarantees rather than a constraint enforced at training time; the unsupervised objective encourages prices that support recovery but does not guarantee the condition. We will add a post-training diagnostic that reports the fraction of instances on which consistency holds. revision: yes

  3. Referee: [Experimental results] Experimental results section: performance tables and generalization plots are reported, but no ablation or diagnostic is included that checks whether the recovered marginals satisfy the feasibility claim or whether the consistency condition is satisfied by the trained prices on the evaluated instances.

    Authors: We will augment the experimental section with new diagnostics that verify global feasibility of recovered marginals and satisfaction of the certificate-consistency condition on both in-distribution and out-of-distribution instances. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims remain conditional and non-reductive

full rationale

The paper's core derivation states that feasibility of the recovered marginal and the first-to-second-order error result hold only when the certificate-consistency condition is satisfied by the neural network's predicted prices. No quoted equation or step in the provided text reduces this condition to a definition in terms of the network outputs themselves, nor does any recovery layer or unsupervised loss term force the consistency by construction. The framework is presented as amortized separation with an explicit external hypothesis, and empirical performance is reported separately from the conditional theory. No self-citation chain, ansatz smuggling, or renaming of known results appears in the load-bearing steps. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on an unelaborated certificate-consistency condition and an asserted local error theory; neural network parameters are implicitly fitted but not enumerated.

free parameters (1)
  • neural network parameters
    Trained unsupervised to predict prices; exact count and regularization unspecified in abstract.
axioms (1)
  • domain assumption certificate-consistency condition holds for predicted prices
    Invoked to guarantee global feasibility of recovered marginal (abstract).

pith-pipeline@v0.9.1-grok · 5705 in / 1123 out tokens · 23547 ms · 2026-07-02T15:09:33.913390+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

51 extracted references · 9 canonical work pages · 3 internal anchors

  1. [1]

    Learning to explore and exploit with gnns for unsupervised combinatorial optimization

    Utku Umur Acikalin, Aaron M Ferber, and Carla P Gomes. Learning to explore and exploit with gnns for unsupervised combinatorial optimization. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  2. [2]

    Differentiable convex optimization layers.Advances in neural information processing systems, 32, 2019

    Akshay Agrawal, Brandon Amos, Shane Barratt, Stephen Boyd, Steven Diamond, and J Zico Kolter. Differentiable convex optimization layers.Advances in neural information processing systems, 32, 2019

    2019

  3. [3]

    Optnet: Differentiable optimization as a layer in neural networks

    Brandon Amos and J Zico Kolter. Optnet: Differentiable optimization as a layer in neural networks. InInternational conference on machine learning, pages 136–145. PMLR, 2017

    2017

  4. [4]

    Deep equilibrium models

    Shaojie Bai, J Zico Kolter, and Vladlen Koltun. Deep equilibrium models. Advances in neural information processing systems, 32, 2019

    2019

  5. [5]

    Emergence of scaling in random networks.science, 286(5439):509–512, 1999

    Albert-László Barabási and Réka Albert. Emergence of scaling in random networks.science, 286(5439):509–512, 1999

    1999

  6. [6]

    Or-library: distributing test problems by electronic mail.Journal of the operational research society, 41(11):1069–1072, 1990

    John E Beasley. Or-library: distributing test problems by electronic mail.Journal of the operational research society, 41(11):1069–1072, 1990

    1990

  7. [7]

    Neural Combinatorial Optimization with Reinforcement Learning

    Irwan Bello, Hieu Pham, Quoc V Le, Mohammad Norouzi, and Samy Bengio. Neural combinatorial optimization with reinforcement learning.arXiv preprint arXiv:1611.09940, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  8. [8]

    Machine learning for combinatorial optimization: a methodological tour d’horizon.European Journal of Operational Research, 290(2):405–421, 2021

    Yoshua Bengio, Andrea Lodi, and Antoine Prouvost. Machine learning for combinatorial optimization: a methodological tour d’horizon.European Journal of Operational Research, 290(2):405–421, 2021

    2021

  9. [9]

    Springer Science & Business Media, 2013

    J Frédéric Bonnans and Alexander Shapiro.Perturbation analysis of optimization problems. Springer Science & Business Media, 2013. 14

    2013

  10. [10]

    Unsupervised learning for the elementary shortest path problem.arXiv preprint arXiv:2508.01557, 2025

    Jingyi Chen, Xinyuan Zhang, and Xinwu Qian. Unsupervised learning for the elementary shortest path problem.arXiv preprint arXiv:2508.01557, 2025

    work page arXiv 2025

  11. [11]

    Neural combinatorial optimization with reinforcement learning in industrial engineering: a survey

    Kwok Tung Chung, Carman KM Lee, and Yung Po Tsang. Neural combinatorial optimization with reinforcement learning in industrial engineering: a survey. Artificial Intelligence Review, 58(5):130, 2025

    2025

  12. [12]

    Large language models for combinatorial optimization: A systematic review.ACM Computing Surveys, 2025

    Francesca Da Ros, Michael Soprano, Luca Di Gaspero, and Kevin Roitero. Large language models for combinatorial optimization: A systematic review.ACM Computing Surveys, 2025

    2025

  13. [13]

    Springer Science & Business Media, 2006

    Guy Desaulniers, Jacques Desrosiers, and Marius M Solomon.Column generation, volume 5. Springer Science & Business Media, 2006

    2006

  14. [14]

    Springer, 2009

    Asen L Dontchev and R Tyrrell Rockafellar.Implicit functions and solution mappings, volume 543. Springer, 2009

    2009

  15. [15]

    On random graphs i.Publ

    P ERDdS and A R&wi. On random graphs i.Publ. math. debrecen, 6(290-297): 18, 1959

    1959

  16. [16]

    Dominique Feillet, Pierre Dejax, Michel Gendreau, and Cyrille Gueguen. An exact algorithm for the elementary shortest path problem with resource constraints: Application to some vehicle routing problems.Networks: An International Journal, 44(3):216–229, 2004

    2004

  17. [17]

    Exact combinatorial optimization with graph convolutional neural networks

    Maxime Gasse, Didier Chételat, Nicola Ferroni, Laurent Charlin, and Andrea Lodi. Exact combinatorial optimization with graph convolutional neural networks. Advances in neural information processing systems, 32, 2019

    2019

  18. [18]

    Geoffrion

    Arthur M. Geoffrion. Lagrangean relaxation for integer programming.Mathe- matical Programming Study, 2:82–114, 1974

    1974

  19. [19]

    Random graphs.The Annals of Mathematical Statistics, 30(4): 1141–1144, 1959

    Edgar N Gilbert. Random graphs.The Annals of Mathematical Statistics, 30(4): 1141–1144, 1959

    1959

  20. [20]

    Primal-dual neural algorithmic reasoning.arXiv preprint arXiv:2505.24067, 2025

    Yu He and Ellen Vitercik. Primal-dual neural algorithmic reasoning.arXiv preprint arXiv:2505.24067, 2025

    work page arXiv 2025

  21. [21]

    An efficient graph convo- lutional network technique for the travelling salesman problem,

    Chaitanya K Joshi, Thomas Laurent, and Xavier Bresson. An efficient graph convolutional network technique for the travelling salesman problem.arXiv preprint arXiv:1906.01227, 2019

    work page arXiv 1906

  22. [22]

    Snap datasets: Stanford large network dataset collection.Retrieved December 2021 from http://snap

    Leskovec Jure. Snap datasets: Stanford large network dataset collection.Retrieved December 2021 from http://snap. stanford. edu/data, 2014. 15

    2021

  23. [23]

    Erdos goes neural: an unsupervised learning framework for combinatorial optimization on graphs.Advances in Neural Information Processing Systems, 33:6659–6672, 2020

    Nikolaos Karalias and Andreas Loukas. Erdos goes neural: an unsupervised learning framework for combinatorial optimization on graphs.Advances in Neural Information Processing Systems, 33:6659–6672, 2020

    2020

  24. [24]

    The cutting-plane method for solving convex programs

    James E Kelley, Jr. The cutting-plane method for solving convex programs. Journal of the society for Industrial and Applied Mathematics, 8(4):703–712, 1960

    1960

  25. [25]

    Kriege, Christopher Morris, Petra Mutzel, and Marion Neumann

    Kristian Kersting, Nils M. Kriege, Christopher Morris, Petra Mutzel, and Marion Neumann. Benchmark data sets for graph kernels.http://www.graphlearning. io/, 2020

    2020

  26. [26]

    Learn- ing combinatorial optimization algorithms over graphs.Advances in neural information processing systems, 30, 2017

    Elias Khalil, Hanjun Dai, Yuyu Zhang, Bistra Dilkina, and Le Song. Learn- ing combinatorial optimization algorithms over graphs.Advances in neural information processing systems, 30, 2017

    2017

  27. [27]

    Attention, Learn to Solve Routing Problems!

    Wouter Kool, Herke Van Hoof, and Max Welling. Attention, learn to solve routing problems!arXiv preprint arXiv:1803.08475, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  28. [28]

    Springer, 2002

    Steven George Krantz and Harold R Parks.The implicit function theorem: history, theory, and applications, volume 202. Springer, 2002

    2002

  29. [29]

    Pomo: Policy optimization with multiple optima for reinforcement learning.Advances in neural information processing systems, 33:21188–21198, 2020

    Yeong-Dae Kwon, Jinho Choo, Byoungjip Kim, Iljoo Yoon, Youngjune Gwon, and Seungjai Min. Pomo: Policy optimization with multiple optima for reinforcement learning.Advances in neural information processing systems, 33:21188–21198, 2020

    2020

  30. [30]

    Finding near-optimal independent sets at scale

    Sebastian Lamm, Peter Sanders, Christian Schulz, Darren Strash, and Renato F Werneck. Finding near-optimal independent sets at scale. In2016 Proceedings of the eighteenth workshop on algorithm engineering and experiments (ALENEX), pages 138–150. SIAM, 2016

    2016

  31. [31]

    Branch-and-bound methods: A survey

    Eugene L Lawler and David E Wood. Branch-and-bound methods: A survey. Operations research, 14(4):699–719, 1966

    1966

  32. [32]

    Yang Li, Jinpei Guo, Runzhong Wang, Hongyuan Zha, and Junchi Yan. Fast t2t: Optimization consistency speeds up diffusion-based training-to-testing solving for combinatorial optimization.Advances in Neural Information Processing Systems, 37:30179–30206, 2024

    2024

  33. [33]

    Combinatorial optimization with graph convolutional networks and guided tree search.Advances in neural information processing systems, 31, 2018

    Zhuwen Li, Qifeng Chen, and Vladlen Koltun. Combinatorial optimization with graph convolutional networks and guided tree search.Advances in neural information processing systems, 31, 2018

    2018

  34. [34]

    Bruce T Lowerre.The harpy speech recognition system.Carnegie Mellon Univer- sity, 1976. 16

    1976

  35. [35]

    Cambridge University Press, 1996

    Zhi-Quan Luo, Jong-Shi Pang, and Daniel Ralph.Mathematical programs with equilibrium constraints. Cambridge University Press, 1996

    1996

  36. [36]

    Sparsemap: Differentiable sparse structured inference

    Vlad Niculae, Andre Martins, Mathieu Blondel, and Claire Cardie. Sparsemap: Differentiable sparse structured inference. InInternational Conference on Ma- chine Learning, pages 3799–3808. PMLR, 2018

    2018

  37. [37]

    Strongly regular generalized equations.Mathematics of Operations Research, 5(1):43–62, 1980

    Stephen M Robinson. Strongly regular generalized equations.Mathematics of Operations Research, 5(1):43–62, 1980

    1980

  38. [38]

    A diffusion model framework for unsupervised neural combinatorial optimization.arXiv preprint arXiv:2406.01661, 2024

    Sebastian Sanokowski, Sepp Hochreiter, and Sebastian Lehner. A diffusion model framework for unsupervised neural combinatorial optimization.arXiv preprint arXiv:2406.01661, 2024

    work page arXiv 2024

  39. [39]

    Combinatorial optimization augmented machine learning.arXiv preprint arXiv:2601.10583, 2026

    Maximilian Schiffer, Heiko Hoppe, Yue Su, Louis Bouvier, and Axel Parmen- tier. Combinatorial optimization augmented machine learning.arXiv preprint arXiv:2601.10583, 2026

    work page arXiv 2026

  40. [40]

    Graph neural networks for job shop scheduling problems: A survey.Computers & Operations Research, 176:106914, 2025

    Igor G Smit, Jianan Zhou, Robbert Reijnen, Yaoxin Wu, Jian Chen, Cong Zhang, Zaharah Bukhsh, Yingqian Zhang, and Wim Nuijten. Graph neural networks for job shop scheduling problems: A survey.Computers & Operations Research, 176:106914, 2025

    2025

  41. [41]

    Tightening LP Relaxations for MAP using Message Passing

    David Sontag, Talya Meltzer, Amir Globerson, Tommi S Jaakkola, and Yair Weiss. Tightening lp relaxations for map using message passing.arXiv preprint arXiv:1206.3288, 2012

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  42. [42]

    Difusco: Graph-based diffusion solvers for combinatorial optimization.Advances in neural information processing systems, 36:3706–3731, 2023

    Zhiqing Sun and Yiming Yang. Difusco: Graph-based diffusion solvers for combinatorial optimization.Advances in neural information processing systems, 36:3706–3731, 2023

    2023

  43. [43]

    Run-csp: unsu- pervised learning of message passing networks for binary constraint satisfaction problems.CoRR, abs/1909.08387, 2019

    Jan Toenshoff, Martin Ritzert, Hinrikus Wolf, and Martin Grohe. Run-csp: unsu- pervised learning of message passing networks for binary constraint satisfaction problems.CoRR, abs/1909.08387, 2019

    work page arXiv 1909

  44. [44]

    Pointer networks.Advances in neural information processing systems, 28, 2015

    Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. Pointer networks.Advances in neural information processing systems, 28, 2015

    2015

  45. [45]

    Graphical models, exponential families, and variational inference.Foundations and Trends®in Machine Learning, 1(1-2):1–305, 2008

    Martin J Wainwright and Michael I Jordan. Graphical models, exponential families, and variational inference.Foundations and Trends®in Machine Learning, 1(1-2):1–305, 2008

    2008

  46. [46]

    A new class of upper bounds on the log partition function.IEEE Transactions on Information Theory, 51(7):2313–2335, 2005

    Martin J Wainwright, Tommi S Jaakkola, and Alan S Willsky. A new class of upper bounds on the log partition function.IEEE Transactions on Information Theory, 51(7):2313–2335, 2005. 17

    2005

  47. [47]

    Satnet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver

    Po-Wei Wang, Priya Donti, Bryan Wilder, and Zico Kolter. Satnet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver. In International Conference on Machine Learning, pages 6545–6554. PMLR, 2019

    2019

  48. [48]

    A bi-level framework for learning to solve combinatorial optimization on graphs.Advances in neural information processing systems, 34:21453–21466, 2021

    Runzhong Wang, Zhigang Hua, Gan Liu, Jiayi Zhang, Junchi Yan, Feng Qi, Shuang Yang, Jun Zhou, and Xiaokang Yang. A bi-level framework for learning to solve combinatorial optimization on graphs.Advances in neural information processing systems, 34:21453–21466, 2021

    2021

  49. [49]

    Deep graph kernels

    Pinar Yanardag and Svn VN Vishwanathan. Deep graph kernels. InProceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining, pages 1365–1374, 2015

    2015

  50. [50]

    Constructing free- energy approximations and generalized belief propagation algorithms.IEEE Transactions on information theory, 51(7):2282–2312, 2005

    Jonathan S Yedidia, William T Freeman, and Yair Weiss. Constructing free- energy approximations and generalized belief propagation algorithms.IEEE Transactions on information theory, 51(7):2282–2312, 2005

    2005

  51. [51]

    Learning for routing: A guided review of recent developments and future directions.Transportation Research Part E: Logistics and Transportation Review, 202:104278, 2025

    Fangting Zhou, Attila Lischka, Balázs Kulcsár, Jiaming Wu, Morteza Haghir Chehreghani, and Gilbert Laporte. Learning for routing: A guided review of recent developments and future directions.Transportation Research Part E: Logistics and Transportation Review, 202:104278, 2025. 18 A Derivation of the Variational Outer Objective Forτ >0, define the Gibbs di...

    2025