Pith. sign in

REVIEW 3 major objections 4 minor 61 references

Neural Embedded Mixed-Integer Optimization for Location-Routing Problems

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A neural cost surrogate embedded in a MIP can make location-allocation decisions for large capacitated location-routing instances within about 1% of best-known solutions in seconds.

desk verdict A useful neural-embedded MIP framework for CLRP whose headline generalization claim is undermined by training-test overlap on the Prodhon benchmark. read the letter →

arxiv 2412.05665 v3 pith:CNO3YCE4 submitted 2024-12-07 math.OC

Waquar Kaleem , Doyoung Lee , Changhyun Kwon , Anirudh Subramanyam This is my paper · ORCID
classification math.OC MSC 90C1190C2790B06
keywords location-routingcapacitatedproblemvehicleroutingmixed-integeroptimizationneuralnetworksurrogatepermutation-invariantsetfunctionlocation-allocationpretrainedmodel
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 is trying to establish that the capacitated location-routing problem (CLRP)—where one must choose depots to open and routes to serve customers—can be tackled by replacing the expensive vehicle-routing subproblem with a learned surrogate cost. It trains a permutation-invariant neural network on vehicle-routing instances to predict, for any depot and any subset of customers, the optimal routing cost, then embeds that network directly into a mixed-integer program whose solution gives the depot-opening and customer-assignment decisions. After a one-time training cost of about two hours, the method provides location-allocation decisions within 1% of the best-known solutions for 200-customer instances, in about 3.5 seconds on average, while remaining a simple, modular pipeline that an off-the-shelf MIP solver can handle. The paper is careful to note that it outputs locations and allocations, not the actual routes; those are computed afterwards with a standard vehicle-routing solver.

What carries the argument

The load-bearing object is the sum-decomposition identity for routing cost: the optimal vehicle-routing cost of a set of customers served by a depot is representable, up to normalization, as $\rho$ applied to the sum of per-customer embeddings $\varphi(s_{ij})$, with normalized features $s_{ij}=(P^{-1}(u_j-u_0^i),P^{-1}(v_j-v_0^i),Q^{-1}d_j)$. This converts a decision-dependent routing cost into a linear function of the assignment variables $x_{ij}$ once $\varphi$ is precomputed, and makes the whole surrogate embeddable as mixed-integer linear constraints through the standard ReLU-network encoding. The design also lets one $\varphi/\rho$ pair serve every depot and any subset size, which is why a single pretrained network can be reused across instances.

What would settle it

Hold out the six 200-customer benchmark instances entirely during training, train the same NEO-LRP pipeline on generic vehicle-routing instances or on subsamples from the other 24 instances, and measure the average optimization gap on the held-out instances; if the average gap is far above the reported 0.80%, the out-of-the-box generalization claim is falsified, whereas a gap near 1% would support it.

Watch

Extended reading notes

Core claim

The paper's central claim is that the difficult part of the capacitated location-routing problem can be pushed into a learned, permutation-invariant cost model. Writing $g_i(S)$ for the optimal vehicle-routing cost of serving a customer set $S$ from depot $i$, the paper proves (Theorem 1) that after normalizing customer coordinates relative to the depot and demands by vehicle capacity, this set function admits a depot-independent decomposition $g_i(S)=\rho\left(\sum_{j\in S}\varphi(s_{ij})\right)$. The functions $\varphi$ and $\rho$ are approximated by ReLU feed-forward networks; because only $\rho$ is embedded in the MIP and it is ReLU-activated, the trained surrogate becomes a system of linear constraints plus binary neuron variables, solvable by an off-the-shelf MIP solver. On the 30-instance benchmark, the learned surrogate's location-allocation decisions average 4.32%, 3.68%, 1.77%, and 0.80% above best known for 20, 50, 100, and 200 customers, with the 200-customer decisions taking about 3.5 seconds on average.

Load-bearing premise

The load-bearing premise of the experiments is that training the surrogate on 10,000 vehicle-routing instances subsampled from the same 30 benchmark instances used for testing is a valid way to measure out-of-the-box generalization; if that overlap inflates the reported gaps, the principal claim of easy generalization to new instances is not established by this evidence.

Editorial extensions

If this is right

  • When the paper is right, a single pretrained surrogate can be reused out of the box on a new CLRP instance of any size, without retraining for that instance.
  • Side constraints such as incompatible customer pairs or depot-specific customer restrictions reduce to adding linear inequalities to the MIP, so the framework extends without new routing logic.
  • The final evaluation is exact: after the MIP chooses depots and assignments, a VRP solver computes true routes and costs, so the reported gaps are against actual routing costs, not surrogate predictions.
  • For 200-customer instances the method beats the tailored heuristic baseline in both average gap (0.80% vs 1.24%) and total time; for 100-customer instances it is much faster but slightly less accurate.
  • Training-data quality matters more than volume: increasing training samples from 100 to 100,000 barely moves the optimization gap, while switching from generic VRP samples to benchmark-subsampled training samples halves it.

Reading between the lines

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

  • The framework's separation of location-allocation from routing means its practical ceiling is the quality of the learned cost ranking, not the accuracy of absolute cost predictions; a ranking-preserving loss could therefore improve the optimization gap more than reducing mean squared error, a direction the paper itself flags for future work.
  • Because the decomposition is depot-independent and feature-based, the same surrogate architecture could be lifted to other integrated problems with set-cost terms, such as two-echelon location-routing or combined production-distribution planning, by retraining the cost model on the corresponding vehicle-routing variant.
  • A natural stress test beyond the paper's experiments would be to hold out entire benchmark instances during training and evaluate on those held-out instances; this would directly measure whether the reported gaps reflect genuine out-of-the-box generalization or overlap between training subsamples and test instances.
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

3 major / 4 minor

Summary. The paper proposes NEO-LRP, a framework that approximates the optimal vehicle routing cost of serving any customer subset from a depot by a permutation-invariant neural network (a Deep Sets architecture), embeds this ReLU network as a mixed-integer surrogate in a CLRP location-allocation MIP, solves the MIP with an off-the-shelf solver, and then evaluates the true routing cost of the resulting allocation with an exact VRP solver. The authors report experiments on the Prodhon benchmark set, comparing against GRASP/VNS and an FLP-VRP baseline, and provide ablations on sampling methods, sample size, routing label solver, problem size, and single versus customized networks. The central technical claims are that the surrogate MIP is easy to solve, that a single pre-trained model can be used out of the box, and that NEO-LRP achieves high-quality location-allocation decisions for large instances in seconds.

Significance. If the generalization claim holds, NEO-LRP offers a genuinely modular and simple alternative to routing-specific heuristics: it requires no routing-specific algorithm engineering, handles side constraints within the MIP, and leverages mature MIP and VRP solvers. The paper has notable strengths: Theorem 1 is a sound application of an existing set-function representation theorem; the final location-allocation solution is evaluated by an exact VRP solver, so the reported objective values are not based on the surrogate's own predictions; the ablation studies are broad and informative; and the authors release code and data. However, the main evidence for out-of-the-box generalization is currently compromised by the RSCC sampling procedure, which draws training VRP instances from the very Prodhon CLRP instances used for testing. The paper's own GVS ablation shows that changing the sampling method substantially changes the optimization gap, so this issue is load-bearing for the headline claims.

major comments (3)
  1. [Abstract and Section 4.4; Appendix B, Algorithm 1] The abstract states that the neural network is trained on a dataset 'entirely separate from any CLRP test instances', but this is contradicted by the experimental protocol. Section 5.4 states that the 10,000 training instances are generated 'across all 30 CLRP instances using the RSCC data sampling method', and Algorithm 1 in Appendix B explicitly selects a CLRP instance from the benchmark set and subsamples customers and a depot from that instance. Consequently, the training VRP instances share depot coordinates, customer coordinates, demands, capacities, and vehicle parameters with the test CLRP instances. The headline result of a 0.80% average gap for 200-customer instances in Table 1 is therefore not evidence of generalization to unseen CLRP instances. I request either retraining with a data-generation procedure that is independent of the Prodhon set (e.g., GVS) and reporting the corresponding Table 1, or substantially revising the abstract and the out-of-the-box generalization claims.
  2. [Section 5.5.1, Figure 3] The ablation on sampling methods shows that the choice of sampling is highly consequential: RSCC achieves a median optimization gap of 1.84%, while GVS, which is independent of the Prodhon instances, achieves 4.29%. The paper attributes this difference to 'alignment between the training data and the customer assignments Ji(x) during optimization', but an equally natural explanation is that RSCC training instances are subsampled from the same benchmark instances used for testing. Because the main experiments in Table 1 use RSCC, the reported gaps may be optimistic for unseen instances. This does not invalidate the framework, but it means the paper currently establishes in-distribution performance on Prodhon rather than out-of-the-box generalization.
  3. [Section 5.5.5, Table 2] The single-versus-customized comparison does not resolve the generalization question. Both the 'single' and 'customized' networks are trained with RSCC on the same Prodhon benchmark set, so the comparison measures the effect of aggregating training data across instances, not the effect of testing on instances that were never used in training. To support the claim that a single pre-trained model generalizes across CLRP instances, the authors should evaluate a model trained without any instances from the test benchmark, or on a separate benchmark family.
minor comments (4)
  1. [Section 4.2] In Section 4.2, the normalization constant P is defined as the maximum range of customer coordinates after centering the depot at (0,0), whereas Theorem 1 treats P as any fixed positive constant. Please clarify that the specific choice of P in the implementation is one admissible instantiation of the theorem's constant.
  2. [Appendix A, Eq. (17)] Equation (17) uses the index m both as the output of the final layer and in the notation w^K_{\ell m}; please adjust the notation to avoid confusion with the hidden-layer index m used in equations (14)-(16).
  3. [Table 2] In the row for instance 200-10-3b in Table 2, the BKS value is printed as '362, 320' with an extra space; please correct the formatting.
  4. [Section 5.4] The phrase 'Tone-time' should be 'one-time' or 'T_one-time'; the subscript appears to be a typographical artifact.

Circularity Check

1 steps flagged · score 4.0 of 10

The headline generalization claim is weakened by training/test overlap: RSCC subsamples training VRP instances from the same Prodhon CLRP instances used for evaluation; the mathematical derivation itself is not circular.

  1. fitted input called prediction [Abstract; Section 4.4 (RSCC); Section 5.4; Appendix B Algorithm 1]
    "The network is trained on an independently generated dataset of CVRP instances from the literature, entirely separate from any CLRP test instances. ... In RSCC, we generate VRP instances by subsampling a depot and customer subsets from a given benchmark dataset of CLRP instances. ... we generated 10,000 unique VRP instances across all 30 CLRP instances using the RSCC data sampling method."

    Algorithm 1 takes the dataset of CLRP instances as input, draws one instance uniformly at random, and within it randomly selects a customer subset and depot. Since the same 30 Prodhon instances are the test bed in Table 1, the neural surrogate is trained on cost-to-geometry mappings drawn from the very instances whose location-allocation decisions are later reported as predictions. The abstract's 'entirely separate' claim is therefore not supported by the actual data-generation procedure. The paper's own ablation makes the leak consequential: RSCC yields median gap 1.84% versus 4.29% for GVS, which is genuinely independent of the Prodhon set. Thus the 'within 1%' large-instance result is partly an in-distribution evaluation, not proof of out-of-the-box generalization.

full rationale

No derivation-level circularity is present in the mathematical core. Theorem 1's sum-decomposition of the VRP cost function is imported from an external representation theorem [60], and the ReLU-network-to-MIP embedding follows the independent formulations of [25] and [27]. The CLRP objective (2)-(5) is not defined in terms of the trained surrogate; rather, the surrogate is embedded to guide location-allocation decisions and final routes are recomputed exactly with VRPSolverEasy, so the final gap is not forced to equal the surrogate output. The only significant circularity signal is empirical: the RSCC sampling procedure used for the main experiments in Section 5.4 constructs training VRP instances by subsampling the same Prodhon CLRP benchmark instances that are then used for testing, contradicting the abstract's claim of a training set 'entirely separate from any CLRP test instances.' The ablation in Section 5.5.1 confirms that this choice is consequential (RSCC 1.84% vs. independent GVS 4.29% median gap). Because the MIP optimization and exact post-hoc routing evaluation still contribute independent content, the paper is not wholly circular, but the headline 'within 1% for large problems' claim is not established as an out-of-the-box generalization result.

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

The central claim rests on the universal set-function representation theorem, the norm-induced travel cost assumption, and the learned surrogate's accuracy. No new physical or mathematical entities are introduced; the neural surrogate is a standard function approximator.

free parameters (3)
  • Neural network weights of phi and rho = Not explicitly reported; learned from 10,000 labeled VRP instances
    The surrogate routing cost is entirely determined by these learned weights, and the central claim depends on their approximation quality.
  • Latent dimension L = Chosen from {4, 6, 8} by hyperparameter optimization
    L controls the expressiveness of the Deep Sets embedding and the size of the embedded MIP; it is selected per run and affects solution quality.
  • Hidden layer counts and neuron counts for phi and rho = phi: 2 to 6 layers with 32 to 2048 neurons; rho: 1 layer with 4 to 8 neurons
    The architecture is chosen by random hyperparameter search, and the MIP complexity and approximation error depend on these choices.
assumptions (3)
  • standard math Universal set-function representation theorem from Wagstaff et al. [60, Theorem 2.8]: any permutation-invariant set function can be written as rho(sum of phi over elements).
    Theorem 1 in the paper relies on this external result to justify the Deep Sets architecture for the optimal VRP cost function.
  • domain assumption Travel costs are norm-induced, so translating the depot to the origin and scaling coordinates by P preserves the optimal VRP cost up to a constant factor.
    This assumption is stated in Section 3 and used in the proof of Theorem 1 to justify depot-independent normalized features.
  • domain assumption The trained neural network approximates the optimal VRP cost well enough on the support of customer subsets that appear in MIP-optimal assignments.
    The whole framework replaces gi with a learned surrogate; no approximation guarantee is proven, so the method's effectiveness rests on this empirical assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Embedded Mixed-Integer Optimization for Location-Routing Problems." pith.science (2026). https://pith.science/paper/CNO3YCE4

@misc{pith2026241205665,
  author       = {Pith},
  title        = {Pith review of: Neural Embedded Mixed-Integer Optimization for Location-Routing Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CNO3YCE4}},
  note         = {Machine review of arXiv:2412.05665}
}
read the original abstract

We present a framework that combines machine learning with mixed-integer optimization to solve the Capacitated Location-Routing Problem (CLRP), a classical NP-hard problem that integrates strategic facility location with operational vehicle routing decisions. The proposed method trains a neural network to approximate the cost of a Capacitated Vehicle Routing Problem (CVRP) for serving any subset of customers from a candidate facility. The network is trained on an independently generated dataset of CVRP instances from the literature, entirely separate from any CLRP test instances, thereby avoiding the overfitting and information leakage that can affect learning-based methods. The trained network is then embedded as a surrogate within a mixed-integer model for location-allocation decisions, which is solved using off-the-shelf solvers, thus leveraging decades of advances in vehicle routing and the availability of mature solvers. Computational experiments across four benchmark sets show that the method delivers reasonable solution quality and scales well to large instances, where, after a one-time training cost, it reaches solutions close to the best known at a fraction of the runtime of state-of-the-art heuristics. Our results demonstrate the value of routing cost approximations from the neural surrogate in informing high-quality location-allocation decisions. Our code and data are publicly available.

Figures

Figures reproduced from arXiv: 2412.05665 by the authors.

Figure 1
Figure 1. An illustrative example of a CLRP solution. Depots and customers are depicted by [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. The NEO-LRP framework for solving the CLRP. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Comparison of sampling methods on (a) Etest, (b) Epred, (c) E gap BKS The GVS sampling method generates instances inspired by [58, 49] and aims to create diverse spatial and demand distributions of customer sets S. This diversity allows GVS to achieve a relatively low median Etest of 2.87%. However, the median Epred is much higher at 58.31%, in￾dicating poor generalization to the customer assignments Ji(x) encounter… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Effect of sample size on (a) Etest, (b) Epred, (c) E gap BKS. ν heuristic(D) and an exact solver [22] with a time limit of 1800 seconds to compute ν exact(D). To quantify the difference between these two solvers, we define the label gap (Elabel) as the percentage discr…
Figure 5
Figure 5. Figure 5: Comparison of exact versus heuristic VRP solvers: (a) Solver time (s) , (b) [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Effect of problem size (the x-axis represents the number of customers in the instance). 5.5.5 Single versus Customized Neural Networks We study the effect of using a single neural network trained on a heterogeneous dataset across all benchmark instances compared to usi…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 57 canonical work pages

  1. [1]

    In: Operations research and cyber-infrastructure

    Akca, Z., Berger, R., Ralphs, T.: A branch-and-price algorithm for combined location and routing problems under capacity restrictions. In: Operations research and cyber-infrastructure. pp. 309–330. Springer (2009)

    work page 2009

  2. [2]

    Lo- cation science pp

    Albareda-Sambola, M., Rodr ´ ıguez-Pereira, J.: Location-routing and location-arc routing. Lo- cation science pp. 431–451 (2019)

    work page 2019

  3. [3]

    In: Laporte, G., Nickel, S., Saldanha-da Gama, F

    Albareda-Sambola, M., Rodr ´ ıguez-Pereira, J.: Location-Routing and Location-Arc Routing. In: Laporte, G., Nickel, S., Saldanha-da Gama, F. (eds.) Location Science, pp. 431–451. Springer, Cham, Switzerland, 2 edn. (2020)

    work page 2020

  4. [4]

    EURO Journal on Transportation and Logistics 2(3), 231–254 (2013)

    Alvim, A.C., Taillard, ´E.D.: Popmusic for the world location-routing problem. EURO Journal on Transportation and Logistics 2(3), 231–254 (2013)

    work page 2013

  5. [5]

    Operations research 59(5), 1284–1296 (2011)

    Baldacci, R., Mingozzi, A., Wolfler Calvo, R.: An exact method for the capacitated location- routing problem. Operations research 59(5), 1284–1296 (2011)

    work page 2011

  6. [6]

    Computers & Operations Research 38(6), 931– 941 (2011)

    Belenguer, J.M., Benavent, E., Prins, C., Prodhon, C., Calvo, R.W.: A branch-and-cut method for the capacitated location-routing problem. Computers & Operations Research 38(6), 931– 941 (2011)

    work page 2011

  7. [7]

    European Journal of Operational Research 290(2), 405–421 (2021)

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

    work page 2021

  8. [8]

    Transportation Science 41(1), 29–43 (2007)

    Berger, R.T., Coullard, C.R., Daskin, M.S.: Location-routing problems with distance con- straints. Transportation Science 41(1), 29–43 (2007)

    work page 2007

Show all 61 references
  1. [9]

    arXiv preprint arXiv:2205.02453 (2022)

    Bogyrbayeva, A., Meraliyev, M., Mustakhov, T., Dauletbayev, B.: Learning to solve vehicle routing problems: A survey. arXiv preprint arXiv:2205.02453 (2022)

    2022 arXiv

  2. [10]

    Mathematical programming 151, 191–223 (2015)

    Bonami, P., Lodi, A., Tramontani, A., Wiese, S.: On mathematical programming with indica- tor constraints. Mathematical programming 151, 191–223 (2015)

    2015

  3. [11]

    In: Knowledge-Based Intelligent Infor- mation and Engineering Systems: 10th International Conference, KES 2006, Bournemouth, UK, October 9-11, 2006

    Bouhafs, L., Hajjam, A., Koukam, A.: A combination of simulated annealing and ant colony system for the capacitated location-routing problem. In: Knowledge-Based Intelligent Infor- mation and Engineering Systems: 10th International Conference, KES 2006, Bournemouth, UK, Octobe...

    2006

  4. [12]

    Journal of Machine Learning Research 24(130), 1–61 (2023)

    Cappart, Q., Ch´ etelat, D., Khalil, E.B., Lodi, A., Morris, C., Veliˇ ckovi´ c, P.: Combinatorial optimization and reasoning with graph neural networks. Journal of Machine Learning Research 24(130), 1–61 (2023)

    2023

  5. [13]

    The Journal of Machine Learning Research 23(1), 15829–15836 (2022)

    Ceccon, F., Jalving, J., Haddad, J., Thebelt, A., Tsay, C., Laird, C.D., Misener, R.: Omlt: Optimization & machine learning toolkit. The Journal of Machine Learning Research 23(1), 15829–15836 (2022)

    2022

  6. [14]

    Mathematical and Com- puter Modelling 41(8-9), 1035–1053 (2005) 24

    Chan, Y., Baker, S.F.: The multiple depot, multiple traveling salesmen facility-location prob- lem: Vehicle range, service frequency, and heuristic implementations. Mathematical and Com- puter Modelling 41(8-9), 1035–1053 (2005) 24

    2005

  7. [15]

    Discrete Optimization 10(4), 263–295 (2013)

    Contardo, C., Cordeau, J.F., Gendron, B.: A computational comparison of flow formulations for the capacitated location-routing problem. Discrete Optimization 10(4), 263–295 (2013)

    2013

  8. [16]

    INFORMS Journal on Computing 26(1), 88–102 (2014)

    Contardo, C., Cordeau, J.F., Gendron, B.: An exact algorithm based on cut-and-column generation for the capacitated location-routing problem. INFORMS Journal on Computing 26(1), 88–102 (2014)

    2014

  9. [17]

    Com- puters & Operations Research 55, 185–199 (2015)

    Cuda, R., Guastaroba, G., Speranza, M.G.: A survey on two-echelon routing problems. Com- puters & Operations Research 55, 185–199 (2015)

    2015

  10. [18]

    Expert Systems with Applications 39(3), 2865– 2871 (2012)

    Derbel, H., Jarboui, B., Hanafi, S., Chabchoub, H.: Genetic algorithm with iterated local search for solving a location-routing problem. Expert Systems with Applications 39(3), 2865– 2871 (2012)

    2012

  11. [19]

    European journal of operational research 241(2), 283–308 (2015)

    Drexl, M., Schneider, M.: A survey of variants and extensions of the location-routing problem. European journal of operational research 241(2), 283–308 (2015)

    2015

  12. [20]

    In: Proceedings of the 9th EU/Meeting on Metaheuristics for Logistics and Vehicle Routing, Troyes, France

    Duhamel, C., Lacomme, P., Prins, C., Prodhon, C.: A memetic approach for the capaci- tated location routing problem. In: Proceedings of the 9th EU/Meeting on Metaheuristics for Logistics and Vehicle Routing, Troyes, France. vol. 38, p. 39 (2008)

    2008

  13. [21]

    Computers & Operations Research 37(11), 1912–1923 (2010)

    Duhamel, C., Lacomme, P., Prins, C., Prodhon, C.: A grasp × els approach for the capacitated location-routing problem. Computers & Operations Research 37(11), 1912–1923 (2010)

    2010

  14. [22]

    INFORMS Journal on Computing 36(4), 956–965 (2024)

    Errami, N., Queiroga, E., Sadykov, R., Uchoa, E.: Vrpsolvereasy: a python library for the exact solution of a rich vehicle routing problem. INFORMS Journal on Computing 36(4), 956–965 (2024)

    2024

  15. [23]

    Computers & Operations Research 40(1), 70–79 (2013)

    Escobar, J.W., Linfati, R., Toth, P.: A two-phase hybrid heuristic algorithm for the capacitated location-routing problem. Computers & Operations Research 40(1), 70–79 (2013)

    2013

  16. [24]

    Springer Science & Business Media (2009)

    Farahani, R.Z., Hekmatfar, M.: Facility location: concepts, models, algorithms and case stud- ies. Springer Science & Business Media (2009)

    2009

  17. [25]

    Constraints 23(3), 296–309 (2018)

    Fischetti, M., Jo, J.: Deep neural networks and mixed integer linear optimization. Constraints 23(3), 296–309 (2018)

    2018

  18. [26]

    Furnon, V., Perron, L.: Or-tools routing library, https://developers.google.com/ optimization/routing/

  19. [27]

    Computers & Chemical Engineering 131, 106580 (2019)

    Grimstad, B., Andersson, H.: Relu networks as surrogate models in mixed-integer linear pro- grams. Computers & Chemical Engineering 131, 106580 (2019)

    2019

  20. [28]

    Gurobi Optimization, L.: gurobi-machinelearning: A python package for mixed-integer pro- gramming formulations of trained machine learning models (2024), https://github.com/ Gurobi/gurobi-machinelearning, python package for embedding trained ML models in MIP optimization

    2024

  21. [29]

    Online (2021), https://www

    Gurobi Optimization, LLC: Gurobi optimizer reference manual. Online (2021), https://www. gurobi.com

    2021

  22. [30]

    International Journal of Industrial Engineering Computations 2(1), 141–154 (2011) 25

    Jabal-Ameli, M., Aryanezhad, M., Ghaffari-Nasab, N.: A variable neighborhood descent based heuristic to solve the capacitated location-routing problem. International Journal of Industrial Engineering Computations 2(1), 141–154 (2011) 25

    2011

  23. [31]

    Computers & Operations Research 40(1), 47–57 (2013)

    Jarboui, B., Derbel, H., Hanafi, S., Mladenovi´ c, N.: Variable neighborhood search for location routing. Computers & Operations Research 40(1), 47–57 (2013)

    2013

  24. [32]

    In: IISE Annual Conference

    Kaleem, W., Ayala, H., Subramanyam, A.: Neural embedded optimization for integrated location and routing problems. In: IISE Annual Conference. Proceedings. pp. 1–6. Institute of Industrial and Systems Engineers (IISE) (2024)

    2024

  25. [33]

    In: Advances in Neural Information Processing Systems

    Khalil, E., Dai, H., Zhang, Y., Dilkina, B., Song, L.: Learning combinatorial optimization algorithms over graphs. In: Advances in Neural Information Processing Systems. vol. 30. Curran Associates, Inc. (2017)

    2017

  26. [34]

    Springer, Cham, Switzer- land, 2 edn

    Laporte, G., Nickel, S., Saldanha-da Gama, F.: Location Science. Springer, Cham, Switzer- land, 2 edn. (2020)

    2020

  27. [35]

    INFORMS Journal on Computing (2023)

    Larsen, E., Frejinger, E., Gendron, B., Lodi, A.: Fast continuous and integer l-shaped heuristics through supervised learning. INFORMS Journal on Computing (2023)

    2023

  28. [36]

    EURO Journal on Computational Optimization 11, 100063 (2023)

    L¨ offler, M., Bartolini, E., Schneider, M.: A conceptually simple algorithm for the capacitated location-routing problem. EURO Journal on Computational Optimization 11, 100063 (2023)

    2023

  29. [37]

    International transactions in operational research 20(6), 795–822 (2013)

    Lopes, R.B., Ferreira, C., Santos, B.S., Barreto, S.: A taxonomical analysis, current methods and objectives on location-routing problems. International transactions in operational research 20(6), 795–822 (2013)

    2013

  30. [38]

    International Transactions in Operational Research 28(6), 2941–2983 (2021)

    Mara, S.T.W., Kuo, R., Asih, A.M.S.: Location-routing problem: a classification of recent research. International Transactions in Operational Research 28(6), 2941–2983 (2021)

    2021

  31. [39]

    European Journal of Operational Research 108(1), 1–15 (1998)

    Min, H., Jayaraman, V., Srivastava, R.: Combined location-routing problems: A synthesis and future research directions. European Journal of Operational Research 108(1), 1–15 (1998)

    1998

  32. [40]

    European journal of oper- ational research 177(2), 649–672 (2007)

    Nagy, G., Salhi, S.: Location-routing: Issues, models and methods. European journal of oper- ational research 177(2), 649–672 (2007)

    2007

  33. [41]

    Extending the horizons: Advances in computing, optimization, and decision technologies pp

    ¨Ozyurt, Z., Aksen, D.: Solving the multi-depot location-routing problem with lagrangian relax- ation. Extending the horizons: Advances in computing, optimization, and decision technologies pp. 125–144 (2007)

    2007

  34. [42]

    Advances in neural information processing systems 32 (2019)

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems 32 (2019)

    2019

  35. [43]

    In: Advances in Neural Information Processing Systems

    Patel, R.M., Dumouchelle, J., Khalil, E., Bodur, M.: Neur2sp: Neural two-stage stochastic programming. In: Advances in Neural Information Processing Systems. vol. 35, pp. 23992– 24005. Curran Associates, Inc. (2022)

    2022

  36. [44]

    In: MOSIM’04 (4` eme Conf

    Prins, C., Prodhon, C., Calvo, R.W.: Nouveaux algorithmes pour le probl` eme de localisation et routage avec contraintes de capacit´ e. In: MOSIM’04 (4` eme Conf. Francophone de Mod´ elisation et Simulation) (2004)

    2004

  37. [45]

    In: European Conference on Evolu- tionary Computation in Combinatorial Optimization

    Prins, C., Prodhon, C., Calvo, R.W.: A memetic algorithm with population management (ma— pm) for the capacitated location-routing problem. In: European Conference on Evolu- tionary Computation in Combinatorial Optimization. pp. 183–194. Springer (2006) 26

    2006

  38. [46]

    4or 4, 221–238 (2006)

    Prins, C., Prodhon, C., Calvo, R.W.: Solving the capacitated location-routing problem by a grasp complemented by a learning process and a path relinking. 4or 4, 221–238 (2006)

    2006

  39. [47]

    Transportation science 41(4), 470–483 (2007)

    Prins, C., Prodhon, C., Ruiz, A., Soriano, P., Wolfler Calvo, R.: Solving the capacitated location-routing problem by a cooperative lagrangean relaxation-granular tabu search heuris- tic. Transportation science 41(4), 470–483 (2007)

    2007

  40. [48]

    European journal of operational research 238(1), 1–17 (2014)

    Prodhon, C., Prins, C.: A survey of recent research on location-routing problems. European journal of operational research 238(1), 1–17 (2014)

    2014

  41. [49]

    In: AAAI-22 workshop on machine learning for operations research (ML4OR) (2021)

    Queiroga, E., Sadykov, R., Uchoa, E., Vidal, T.: 10,000 optimal cvrp solutions for testing machine learning based heuristics. In: AAAI-22 workshop on machine learning for operations research (ML4OR) (2021)

    2021

  42. [50]

    In: XIII Congreso de Ingenier ´ ıa de Organizaci´ on

    Sahraeian, R., Nadizadeh, A.: Using greedy clustering method to solve capacitated location- routing problem. In: XIII Congreso de Ingenier ´ ıa de Organizaci´ on. pp. 1721–1729 (2009)

    2009

  43. [51]

    European journal of operational research 39(2), 150–156 (1989)

    Salhi, S., Rand, G.K.: The effect of ignoring routes when locating depots. European journal of operational research 39(2), 150–156 (1989)

    1989

  44. [52]

    Annals of Op- erations Research 259, 389–414 (2017)

    Schneider, M., Drexl, M.: A survey of the standard location-routing problem. Annals of Op- erations Research 259, 389–414 (2017)

    2017

  45. [53]

    Transportation Science 53(1), 301–318 (2019)

    Schneider, M., L¨ offler, M.: Large composite neighborhoods for the capacitated location-routing problem. Transportation Science 53(1), 301–318 (2019)

    2019

  46. [54]

    Transportation Science (2024)

    Sobhanan, A., Park, J., Park, J., Kwon, C.: Genetic algorithms with neural cost predictor for solving hierarchical vehicle routing problems. Transportation Science (2024)

    2024

  47. [55]

    International journal of production economics 141(1), 34–44 (2013)

    Ting, C.J., Chen, C.H.: A multiple ant colony optimization algorithm for the capacitated location routing problem. International journal of production economics 141(1), 34–44 (2013)

    2013

  48. [56]

    arXiv preprint arXiv:1711.07356 (2017)

    Tjeng, V., Xiao, K., Tedrake, R.: Evaluating robustness of neural networks with mixed integer programming. arXiv preprint arXiv:1711.07356 (2017)

    2017 arXiv

  49. [57]

    SIAM (2014)

    Toth, P., Vigo, D.: Vehicle routing: problems, methods, and applications. SIAM (2014)

    2014

  50. [58]

    European Journal of Operational Research 257(3), 845–858 (2017)

    Uchoa, E., Pecin, D., Pessoa, A., Poggi, M., Vidal, T., Subramanian, A.: New benchmark in- stances for the capacitated vehicle routing problem. European Journal of Operational Research 257(3), 845–858 (2017)

    2017

  51. [59]

    ¨O., Albey, E.: Neural network estimators for optimal tour lengths of trav- eling salesperson problem instances with arbitrary node distributions

    Varol, T., ¨Ozener, O. ¨O., Albey, E.: Neural network estimators for optimal tour lengths of trav- eling salesperson problem instances with arbitrary node distributions. Transportation Science 58(1), 45–66 (2024)

    2024

  52. [60]

    In: International Conference on Machine Learning

    Wagstaff, E., Fuchs, F., Engelcke, M., Posner, I., Osborne, M.A.: On the limitations of repre- senting functions on sets. In: International Conference on Machine Learning. pp. 6487–6494. PMLR (2019)

    2019

  53. [61]

    In: Proceedings of the 31st International Conference on Neural Information Processing Systems

    Zaheer, M., Kottur, S., Ravanbhakhsh, S., P´ oczos, B., Salakhutdinov, R., Smola, A.J.: Deep sets. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. pp. 3394–3404 (2017) 27

    2017

Pith tools

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