Pith. sign in

REVIEW 3 major objections 5 minor 61 references

Not All Problems Are Best Modeled as MILP: A DSL-Centric Framework for Flexible and Accurate Optimization Modeling

T0 review · 3 major / 5 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 This paper argues that automated optimization modeling should translate natural language into domain-specific data files rather than mixed-integer linear programs, and reports large gains in accuracy and speed from doing so.

desk verdict The DSL-centric framing is a genuine contribution worth refereeing, but the 51.66% accuracy claim cannot be interpreted until the DSL-to-MILP conversion behind the Gurobi evaluation is disclosed and the template-aligned benchmark is supplemented with independent descriptions. read the letter →

arxiv 2608.07040 v1 pith:SNHQM6PN submitted 2026-08-07 cs.AI

Shaofeng Zhang , Hongyuan Su , Qingwen Peng , Zefang Zong , Shengcai Liu , Ke Tang , Yong Li This is my paper · ORCID
classification cs.AI
keywords combinatorialoptimizationdomain-specificlanguageMILPLLM-basedmodelingautomatedformulationsolverselectionOptiDSLBenchvehiclerouting
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 is trying to establish that the standard way of automating optimization modeling—having a large language model translate a problem description into a mixed-integer linear program (MILP) and then solve it with a MILP solver—is the wrong default for many combinatorial optimization problems. The proposed alternative, OptiDSL, makes the intermediate representation a domain-specific data file, the kind of standardized format that a problem family's own solvers already read and write. The central empirical claim is that this DSL-centric pipeline beats MILP-based LLM pipelines on a new 44-problem-type benchmark by 51.66% in formulation accuracy (optimality rate) and 91.71% in modeling time, and by 23.09% on an existing benchmark. The intended significance is that the bottleneck in LLM-based optimization is not solver strength but the choice of modeling target: picking the right domain representation avoids constraint explosion and opens the pipeline to heuristics and learning-based solvers.

What carries the argument

The load-bearing object is the DSL template pool: a library of standardized data-file formats (VRPLIB-style files for routing, OR-Library-style files for scheduling, packing, and knapsack, and network files for graph problems) that domain solvers already consume. A two-stage LLM workflow first performs semantic routing, matching the problem description to a template using concise meta-descriptions, then instantiates the template, filling fields and deducing flags like OPEN_ROUTE from the wording. An adaptive solver-execution layer profiles solvers offline and routes each DSL file to an exact, heuristic, or learning-based solver according to user preference.

What would settle it

Take an independently written set of optimization problems from textbooks, industrial case studies, or community forums, with no overlap with OptiDSL's template pool, and run both OptiDSL and a MILP-based baseline on them; if the DSL pipeline's optimality-rate advantage disappears or reverses on these out-of-template descriptions, the central claim is not supported. A cheaper check is to inspect the existing external benchmark used in the paper: it covers only CVRP, JSSP, MIS, and MVC, so a failure on any one of these domains at scale would already pressure the generalization claim.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that text-to-MILP is a self-imposed restriction rather than a necessity. For problems such as capacitated vehicle routing, forcing a description into linear constraints produces an explosion of subtour-elimination constraints that overtaxes an LLM, while the same description maps cleanly onto a VRPLIB-style data file. OptiDSL operationalizes this by having an LLM first select a matching DSL template and then fill it with data and deduced logical flags, so formulation is decoupled from execution. The reported results—optimality rates above 80% across five problem domains, with the largest gains in vehicle routing—are taken as evidence that the DSL representation is the more faithful and tractable target for automated modeling.

Load-bearing premise

The evaluation assumes that the natural-language descriptions in the benchmark, generated from the same DSL templates and placeholder tags that OptiDSL is told to fill, are representative of real-world optimization problems, so the reported accuracy gains largely measure template recovery rather than generalization to genuinely new descriptions.

Editorial extensions

If this is right

  • If the central claim is correct, automated optimization pipelines should represent problems in the data format of the target domain rather than forcing them into a universal MILP schema.
  • The reported gains imply that a large part of LLM modeling error comes from the output representation itself, so the same reasoning model can be much more accurate when given a simpler, structured target.
  • Because DSL files are solver-native, LLM-based automation can exploit specialized solvers—exact MILP for small instances, heuristics and neural solvers for large ones—selecting per scale and time budget.
  • The new benchmark of 44 COP types with 4,400 instances gives the community a larger and more diverse testbed for automated modeling than the existing MILP-centric datasets.
  • Costs drop sharply: shaping output as a compact data file reduces modeling time and token consumption, which matters for deploying such systems in latency-sensitive settings.

Reading between the lines

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

  • An implication the authors leave implicit is that the measured advantage may be partly an output-economics effect: data files require far fewer output tokens than full mathematical models, so the same model solves a cheaper task; separating this from representational fidelity would require a token-matched comparison.
  • We infer the template-pool design should transfer to other structured optimization families (e.g., network design, production planning) as long as a domain-standard data format exists; the framework says little about problems without such a standard.
  • A testable extension is to feed OptiDSL problem descriptions whose data are not given as explicit bracketed lists—for example, textual mentions like 'each customer needs about a third of a truckload'—to see whether the DSL grounding step handles implicit quantities as well as explicit placeholders.
  • Because the benchmark descriptions are generated from the same template pool the system is instructed to use, the headline numbers likely measure template recovery; external, independently written problem sets would reveal how much of the gain transfers to genuinely novel phrasings.
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 / 5 minor

Summary. The paper proposes OptiDSL, a framework that uses LLMs to translate natural-language optimization problem descriptions into domain-specific language (DSL) data files (e.g., VRPLIB, OR-Library formats) instead of MILP formulations. The framework includes a two-stage LLM workflow (semantic routing to a DSL template, then instantiation of the template), an adaptive solver-routing mechanism that selects among exact, heuristic, and learning-based solvers, and a new benchmark, OptiDSLBench, spanning 44 COP types across five domains. The authors report that OptiDSL outperforms MILP-based baselines (CoE, ORLM, LLMOPT) in execution rate (ER), optimality rate (OR), and modeling time on OptiDSLBench, and also reports gains on an existing COP benchmark and on MILP-oriented benchmarks.

Significance. If the empirical claims are correct, the paper makes a useful contribution by demonstrating that a DSL-centric intermediate representation can reduce LLM formulation burden and enable solver flexibility for combinatorial optimization, an alternative to the dominant text-to-MILP paradigm. The construction of a large, multi-domain benchmark (4,400 instances, 44 types) is a valuable asset, and the external-benchmark results provide some independent grounding. However, the headline quantitative claims are currently undermined by two load-bearing evaluation issues: the unspecified DSL-to-MILP conversion used for the unified Gurobi evaluation, and the template-aligned construction of OptiDSLBench. The core idea is defensible, but the evidence as presented does not yet support the strong '51.66% gain' and '91.71% decrease' statements.

major comments (3)
  1. [Experiments, Experimental Setup] The experiments report that Gurobi is used as the 'unified downstream solver across all evaluations,' but the paper never specifies how OptiDSL's DSL data files are converted into a Gurobi MILP model. The DSL formats are data formats (coordinates, demands, capacity), not modeling languages; the objective, decision variables, and constraints (e.g., subtour elimination for CVRP) are absent and must be supplied by an external converter. If that converter hard-codes the mathematical model per problem type, then the reported 51.66% OR gain conflates the LLM's data extraction with a pre-specified modeling layer, and the comparison against MILP baselines is not a comparison of formulation accuracy. The paper must disclose the conversion mechanism and either subtract its modeling content or treat it as part of the proposed pipeline and compare end-to-end fairly.
  2. [COPs Benchmark Evaluation, Data Placeholder Substitution] OptiDSLBench is constructed by generating natural-language descriptions from the same DSL template pool and placeholder tags (e.g., <item_weight>, <demand>) that OptiDSL is explicitly instructed to fill. High ER and OR on this benchmark therefore partly measure template recovery rather than general optimization-modeling ability. The external LLMCoSolver benchmark covers only CVRP, JSSP, MIS, and MVC, which is too narrow to support the headline claim of superiority across 44 COP types. The paper should either evaluate on independently authored problem descriptions that do not share placeholder/template structure, or report a separate breakdown for template-aligned vs. non-template instances.
  3. [Overall Performance Comparison, Table 2] The abstract and the main text state that OptiDSL 'significantly surpasses' MILP pipelines, but no variance information or statistical significance tests are reported. All ER/OR values in Table 2 are point estimates over 100 instances without standard deviations, confidence intervals, or paired tests, and some individual rows show OptiDSL below a baseline (e.g., OVRPMBL: OptiDSL ER 0.78 vs. CoE 0.84). A paired test across instances or across problem types is needed to support the use of 'significantly'; otherwise the claim should be reworded to report observed mean differences.
minor comments (5)
  1. [Table 2] The main table shows only partial VRP results with an ellipsis and defers the remaining 24 types to the supplementary material; since the main text cites average VRP gains of 13.05% ER and 68.83% OR, the aggregate VRP row should be included in the main table to make the claim verifiable.
  2. [Figure 2] The workflow diagram includes 'COPs Benchmark Evaluation' as a component of the framework, but the benchmark is an evaluation artifact rather than part of the online OptiDSL pipeline; the caption or diagram should clarify this distinction.
  3. [Introduction, Figure 1] The text attributes the Dantzig-Fulkerson-Johnson formulation to 'Miller, Tucker, and Zemlin 1960' in the Introduction; the DFJ subtour elimination is due to Dantzig, Fulkerson, and Johnson (1954), while Miller-Tucker-Zemlin is a different (MTZ) formulation. The citation and description should be corrected.
  4. [Tables 2 and 4] The OR values in Table 2 are reported as fractions (e.g., 0.93) while Table 4 reports percentages (e.g., 100.0 (6/6)); the units should be made consistent across tables for readability.
  5. [References] The reference for Blackstone et al. contains a typo: 'and, G. L. H.' should be corrected to the proper author formatting for the Handbook of dispatching rules citation.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline 51.66% accuracy gain is largely template-recovery because OptiDSLBench is generated from the same DSL templates and placeholder tags that OptiDSL is explicitly prompted to fill; external LLMCoSolver results provide only partial independent grounding.

  1. self definitional [COPs Benchmark Evaluation, Data Placeholder Substitution]
    "Because large-scale numerical data generation via LLMs is error-prone, we decouple structural generation from data instantiation. The placeholder tags (e.g., ⟨demand⟩) are explicitly defined by the user. Guided by the problem's semantic context and the provided DSL field descriptions, the LLM automatically identifies and matches these user-defined placeholders during formulation, focusing purely on constraint validity."

    The benchmark's natural-language descriptions are constructed by instantiating user-defined placeholder tags that correspond directly to DSL fields, and the ground-truth DSL file for each instance is the template that generated the description. OptiDSL's task is to map the description back into that same DSL template. Thus ER and OR on OptiDSLBench measure how well the LLM recovers the template instantiation that produced the text, not independent modeling quality. The headline 51.66% gain in formulation accuracy over MILP baselines is therefore partly self-referential: the test set is generated from the same DSL schema that OptiDSL is designed to output, while the MILP baselines must construct the mathematical model without that schema.

  2. self definitional [Supplementary Material, Prompt for DSL formulation]
    "You are an information extraction system. Task: Select the SINGLE matching template and fill it with data. Rules: -Output ONLY the completed template. -Keep field names, order, and structure EXACTLY as in the template. -If a field is missing in the problem, fill in FALSE. ... Templates:"

    The deployed formulation prompt defines the task as selecting a template from a provided pool and filling it with data, i.e., slot filling rather than mathematical modeling from first principles. Since the OptiDSLBench descriptions were generated from the same template pool and placeholder tags, the accuracy numbers on that benchmark reduce to measuring the LLM's ability to invert the dataset generator. The reported 51.66% OR advantage is consequently inflated by giving OptiDSL the answer schema while baselines receive no such schema. The external LLMCoSolver benchmark provides some independent support, but it covers only four problem types, so the central quantified claim rests substantially on the self-referential benchmark.

full rationale

The paper's core idea is not circular in itself: OptiDSL genuinely changes the intermediate representation from MILP to domain-standard DSL files, and the external LLMCoSolver and MILP benchmark results give some independent evidence that the approach works. There is no load-bearing self-citation chain, and the DSL templates are external standards (VRPLIB, OR-Library, etc.). However, the headline empirical claim, a 51.66% gain in formulation accuracy on a 44-type benchmark, is substantially self-referential. The benchmark is built by generating natural-language descriptions from the very DSL template pool and placeholder tags that OptiDSL is prompted to fill, and the supplementary prompt explicitly frames formulation as selecting a single matching template and filling it. Accuracy on OptiDSLBench therefore largely measures template recovery, not general formulation skill. A related evaluation threat is that DSL data files must be converted to MILP for the unified Gurobi comparison, and that conversion is unspecified; this is a correctness and interpretability gap rather than a circular reduction by the paper's own equations, so it is not counted as a separate circular step. Overall, partial circularity exists in the central benchmark, but independent external results prevent the score from rising to 8 or 10.

Assumptions & free parameters 4 free parameters · 6 assumptions · 1 invented entities

The central claim depends on a self-generated benchmark whose descriptions are constructed from the DSL templates and placeholder tags OptiDSL is designed to fill, on an unspecified DSL-to-Gurobi conversion, and on several hand-set instance-generation parameters. These choices are not externally imposed, so the reported effect sizes should be treated as upper bounds until independent benchmarks confirm them.

free parameters (4)
  • Evaluation instance sizes = VRP n=5; other domains n=10
    Chosen so Gurobi can solve every instance to optimality. This hand-set regime shapes the measured optimality rates and may understate the difficulty of realistic-scale formulation.
  • Knapsack capacity fraction = 0.6 of total item weight and volume
    Hand-chosen in the supplementary data-generation section. It changes the feasible region and therefore the optimality rates on KP instances.
  • CVRP capacity schedule = C = 30 + floor(n/5) for 20 < n <= 1000; C = 30 otherwise
    Hand-defined formula used to generate CVRP instances in OptiDSLBench. It influences route feasibility and instance difficulty.
  • Instances per COP type = 100
    The number of test instances per problem type is a design choice; it determines the precision of the reported ER/OR point estimates, but no variance is reported.
assumptions (6)
  • domain assumption The domain-standard DSL templates fully and unambiguously capture the semantics of all 44 COP types.
    Invoked in DSL-Based Task Formulation; if a template cannot express a constraint, the DSL output can silently misrepresent the problem.
  • domain assumption The LLM can correctly route each problem description to the right template from a short meta-description.
    Invoked in DSL semantic routing; routing accuracy is not measured separately, so all downstream accuracy depends on it.
  • domain assumption The LLM-generated descriptions in OptiDSLBench are valid, diverse, and representative of real-world COP statements.
    The benchmark is built by LLM generation plus a checking agent, but no large human-authored validation set is provided.
  • domain assumption The hidden conversion from DSL output to a Gurobi model introduces no formulation errors in the unified evaluation.
    Experimental Setup states Gurobi is the unified solver for OptiDSL, but the DSL-to-MILP conversion is not described or validated.
  • domain assumption Placeholder-based descriptions preserve the full information of the optimization problem.
    Data Placeholder Substitution uses explicit tags such as <demand> that map one-to-one to DSL fields, removing ambiguity that real descriptions contain.
  • domain assumption Optimality rate under Gurobi on tiny instances is a faithful measure of formulation quality.
    OR conflates modeling errors with solver or encoding issues and is measured only at sizes small enough for Gurobi to solve to optimality.
invented entities (1)
  • OptiDSLBench
    purpose: Primary evaluation dataset of 4,400 instances across 44 COP types used to support the headline accuracy and speed claims.
    The benchmark is generated by the authors using the same DSL templates and placeholder conventions the framework targets; it is not an independent external validation signal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Not All Problems Are Best Modeled as MILP: A DSL-Centric Framework for Flexible and Accurate Optimization Modeling." pith.science (2026). https://pith.science/paper/SNHQM6PN

@misc{pith2026260807040,
  author       = {Pith},
  title        = {Pith review of: Not All Problems Are Best Modeled as MILP: A DSL-Centric Framework for Flexible and Accurate Optimization Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SNHQM6PN}},
  note         = {Machine review of arXiv:2608.07040}
}
read the original abstract

Solving combinatorial optimization problems (COPs) requires not only efficient algorithms but also carefully crafted formulations. While recent works have leveraged LLMs to automate optimization modeling, current frameworks predominantly rely on a rigid mixed-integer linear programming (MILP) paradigm. In this paper, we argue that not all problems are best modeled as MILP, as forcing complex domains into linear constraints can induce prohibitive modeling complexity and severely restrict solver flexibility. To address this, we propose OptiDSL, a framework that shifts the focus from rigid MILP formulations to domain-specific language (DSL) representations. By utilizing LLMs to map natural language onto standardized, domain-accepted structures, OptiDSL decouples problem formulation from execution. This paradigm enables seamless integration with a diverse library of specialized solvers, ranging from traditional heuristics to modern learning-based methods. Experimental results on the comprehensive benchmark of 44 COP types show that OptiDSL significantly surpasses MILP-based pipelines, yielding a 51.66% gain in formulation accuracy and a 91.71% decrease in modeling time. Notably, it also outperforms MILP-based pipelines on the existing benchmark, achieving a 23.09% higher formulation accuracy. Our code is available at https://anonymous.4open.science/r/OptiDSL.

Figures

Figures reproduced from arXiv: 2608.07040 by the authors.

Figure 1
Figure 1. An illustrative example of modeling paradigms for the CVRP case. The DSL formulation is more appropriate than the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. OptiDSL Workflow: The framework begins with DSL Task Formulation (left), where an LLM translates unstructured natural language descriptions into standardized DSL data templates. These structured instances are then passed to the Adaptive Solver Execution module (center), which dynamically routes the tasks to the optimal solvers based on performance trade-offs. Finally, the COPs Benchmark Evaluation (right) systematic… view at source ↗
Figure 3
Figure 3. Comparison of modeling time and token consumption across different domains. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Performance of different methods on CVRP across [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 1
Figure 1. Figure 1: Transformation of natural language descriptions into structured data files via LLM [PITH_FULL_IMAGE:figures/full_fig_p012_1.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 47 canonical work pages

  1. [1]

    European Journal of Operational Research , volume=

    Maintenance scheduling in the electricity industry: A literature review , author=. European Journal of Operational Research , volume=. 2016 , publisher=

    work page 2016

  2. [2]

    Proceedings of the NeurIPS 2022 Competitions Track , pages =

    NL4Opt Competition: Formulating Optimization Problems Based on Their Natural Language Descriptions , author =. Proceedings of the NeurIPS 2022 Competitions Track , pages =. 2022 , volume =

    work page 2022

  3. [3]

    Chain-of-Experts: When

    Ziyang Xiao and Dongxiang Zhang and Yangjun Wu and Lilin Xu and Yuan Jessica Wang and Xiongwei Han and Xiaojin Fu and Tao Zhong and Jia Zeng and Mingli Song and Gang Chen , booktitle=. Chain-of-Experts: When

    work page

  4. [4]

    2024 , volume =

    Ahmaditeshnizi, Ali and Gao, Wenzhi and Udell, Madeleine , booktitle =. 2024 , volume =

    work page 2024

  5. [5]

    Operations Research , author =

    ORLM: A Customizable Framework in Training Large Models for Automated Optimization Modeling , issn =. Operations Research , author =

    work page

  6. [6]

    Findings of the Association for Computational Linguistics: NAACL 2025 , year =

    LLMs for Mathematical Modeling: Towards Bridging the Gap between Natural and Mathematical Languages , author =. Findings of the Association for Computational Linguistics: NAACL 2025 , year =

    work page 2025

  7. [7]

    Solomon , title =

    Guy Desaulniers and Jacques Desrosiers and Marius M. Solomon , title =. Column Generation , publisher =. 2005 , pages =

    work page 2005

  8. [8]

    Mathematical Programming Computation , volume =

    Daniel Rehfeldt and Thorsten Koch and Yuji Shinano , title =. Mathematical Programming Computation , volume =

    work page

Show all 61 references
  1. [9]

    Buriol , title =

    Henrique Becker and Luciana S. Buriol , title =. European Journal of Operational Research , volume =

  2. [10]

    Mathematical Programming Computation , volume =

    Tobias Achterberg , title =. Mathematical Programming Computation , volume =

  3. [11]

    arXiv preprint arXiv:2202.02799 , year =

    Hengameh Fakhravar , title =. arXiv preprint arXiv:2202.02799 , year =

    arXiv

  4. [12]

    Avtomatika i Telemekhanika , volume=

    Genetic Engineering Algorithm (GEA): An Efficient Metaheuristic Algorithm for Solving Combinatorial Optimization Problems , author=. Avtomatika i Telemekhanika , volume=

  5. [13]

    International Journal of Shipping and Transport Logistics , volume=

    Solving Vehicle Routing Problem with Simultaneous Pickup and Delivery Using Parallel Simulated Annealing Algorithm , author=. International Journal of Shipping and Transport Logistics , volume=. 2016 , publisher=

    2016

  6. [14]

    Natalia Flerova , title =

  7. [15]

    Computational Optimization and Applications , volume =

    Luca Consolini and Mattia Laurini and Marco Locatelli , title =. Computational Optimization and Applications , volume =

  8. [16]

    ICML 2024 Workshop on Foundation Models in the Wild , year=

    Berto, Federico and Hua, Chuanbo and Zepeda, Nayeli Gast and Hottung, Andr. ICML 2024 Workshop on Foundation Models in the Wild , year=

    2024

  9. [17]

    Proceedings of the 37th Conference on Neural Information Processing Systems (NeurIPS 2023) , year=

    DIFUSCO: Graph-based Diffusion Solvers for Combinatorial Optimization , author=. Proceedings of the 37th Conference on Neural Information Processing Systems (NeurIPS 2023) , year=

    2023

  10. [18]

    arXiv preprint arXiv:2406.13125 , year =

    Yaochu Jin and Xueming Yan and Shiqing Liu and Xiangyu Wang , title =. arXiv preprint arXiv:2406.13125 , year =

    arXiv

  11. [19]

    Proceedings of the International Conference on Learning Representations (ICLR) Workshops , year=

    Neural combinatorial optimization with reinforcement learning , author=. Proceedings of the International Conference on Learning Representations (ICLR) Workshops , year=

  12. [20]

    Advances in Neural Information Processing Systems , volume=

    POMO: Policy Optimization with Multiple Optima for Reinforcement Learning , author=. Advances in Neural Information Processing Systems , volume=

  13. [21]

    2000 , number =

    Keld Helsgaun , title =. 2000 , number =

    2000

  14. [22]

    Wouda and Leon Lan and Wouter Kool , title =

    Niels A. Wouda and Leon Lan and Wouter Kool , title =. 2024 , volume =

    2024

  15. [23]

    International Conference on Learning Representations , year=

    Learning Efficient Online 3D Bin Packing on Packing Configuration Trees , author=. International Conference on Learning Representations , year=

  16. [24]

    Qwen3 , url =

    Qwen Team , month =. Qwen3 , url =

  17. [25]

    arXiv preprint arXiv:2412.15115 , year =

    Qwen2.5 Technical Report , author =. arXiv preprint arXiv:2412.15115 , year =

    arXiv

  18. [26]

    arXiv preprint arXiv:2407.10671 , year =

    Qwen2 Technical Report , author =. arXiv preprint arXiv:2407.10671 , year =

    arXiv

  19. [27]

    arXiv preprint arXiv:2412.19437 , year=

    Deepseek-v3 technical report , author=. arXiv preprint arXiv:2412.19437 , year=

    arXiv

  20. [28]

    arXiv preprint arXiv:2501.12948 , year=

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2501.12948 , year=

    arXiv

  21. [29]

    arXiv preprint arXiv:2407.21783 , year=

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    arXiv

  22. [30]

    Complex System Modeling and Simulation , volume=

    A review of reinforcement learning based intelligent optimization for manufacturing scheduling , author=. Complex System Modeling and Simulation , volume=. 2021 , publisher=

    2021

  23. [31]

    ACM Trans

    Zong, Zefang and Feng, Tao and Wang, Jingwei and Xia, Tong and Li, Yong , title =. ACM Trans. Knowl. Discov. Data , articleno =. 2025 , publisher =

    2025

  24. [32]

    Journal of Machine Learning Research , volume=

    Palm: Scaling language modeling with pathways , author=. Journal of Machine Learning Research , volume=

  25. [33]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

  26. [34]

    2024 , url =

    Gurobi Optimization, LLC , title =. 2024 , url =

    2024

  27. [35]

    2009 , edition =

    IBM ILOG Cplex , title =. 2009 , edition =

    2009

  28. [36]

    Computers & Industrial Engineering , volume=

    The vehicle routing problem: A taxonomic review , author=. Computers & Industrial Engineering , volume=. 2009 , publisher=

    2009

  29. [37]

    International Conference on Learning Representations , year=

    Attention, Learn to Solve Routing Problems! , author=. International Conference on Learning Representations , year=

  30. [38]

    Blackstone and Don T

    John H. Blackstone and Don T. Phillips and Gary L. Hogg and , title =. International Journal of Production Research , volume =. 1982 , publisher =

    1982

  31. [39]

    Learning to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning , volume =

    Zhang, Cong and Song, Wen and Cao, Zhiguang and Zhang, Jie and Tan, Puay Siew and Chi, Xu , booktitle =. Learning to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning , volume =

  32. [40]

    Flexible Job-Shop Scheduling via Graph Neural Network and Deep Reinforcement Learning , year=

    Song, Wen and Chen, Xinyang and Li, Qiqiang and Cao, Zhiguang , journal=. Flexible Job-Shop Scheduling via Graph Neural Network and Deep Reinforcement Learning , year=

  33. [41]

    Flexible Job Shop Scheduling via Dual Attention Network-Based Reinforcement Learning , year=

    Wang, Runqing and Wang, Gang and Sun, Jian and Deng, Fang and Chen, Jie , journal=. Flexible Job Shop Scheduling via Dual Attention Network-Based Reinforcement Learning , year=

  34. [42]

    Matrix encoding networks for neural combinatorial optimization , volume =

    Kwon, Yeong-Dae and Choo, Jinho and Yoon, Iljoo and Park, Minah and Park, Duwon and Gwon, Youngjune , booktitle =. Matrix encoding networks for neural combinatorial optimization , volume =

  35. [43]

    Darko Drakulic and Sofia Michel and Jean-Marc Andreoli , booktitle=

  36. [44]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    Fast T2T: Optimization Consistency Speeds Up Diffusion-Based Training-to-Testing Solving for Combinatorial Optimization , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

  37. [45]

    Proceedings of the 41st International Conference on Machine Learning , pages =

    A Diffusion Model Framework for Unsupervised Neural Combinatorial Optimization , author =. Proceedings of the 41st International Conference on Machine Learning , pages =. 2024 , volume =

    2024

  38. [46]

    arXiv preprint arXiv:2402.02244 , year=

    Beyond the limits: A survey of techniques to extend the context length in large language models , author=. arXiv preprint arXiv:2402.02244 , year=

    arXiv

  39. [47]

    arXiv preprint arXiv:2410.05983 , year=

    Long-context llms meet rag: Overcoming challenges for long inputs in rag , author=. arXiv preprint arXiv:2410.05983 , year=

    arXiv

  40. [48]

    An Extension of the Lin-Kernighan-Helsgaun TSP Solver for Constrained Traveling Salesman and Vehicle Routing Problems: Technical report

    Keld Helsgaun , year =. An Extension of the Lin-Kernighan-Helsgaun TSP Solver for Constrained Traveling Salesman and Vehicle Routing Problems: Technical report

  41. [49]

    1999 , publisher=

    Evolutionary computation: Theory and applications , author=. 1999 , publisher=

    1999

  42. [50]

    ACM Trans

    Huang, Lei and Yu, Weijiang and Ma, Weitao and Zhong, Weihong and Feng, Zhangyin and Wang, Haotian and Chen, Qianglong and Peng, Weihua and Feng, Xiaocheng and Qin, Bing and Liu, Ting , title =. ACM Trans. Inf. Syst. , articleno =. 2025 , issue_date =

    2025

  43. [51]

    Science , volume =

    Albert-László Barabási and Réka Albert , title =. Science , volume =

  44. [52]

    Proceedings of the 19th International Joint Conference on Artificial Intelligence , pages =

    Xu, Ke and Boussemart, Fr´ed´eric and Hemery, Fred and Lecoutre, Christophe , title =. Proceedings of the 19th International Joint Conference on Artificial Intelligence , pages =. 2005 , publisher =

    2005

  45. [53]

    Miller, C. E. and Tucker, A. W. and Zemlin, R. A. , title =. Journal of the ACM , pages =. 1960 , issue_date =

    1960

  46. [54]

    The Thirteenth International Conference on Learning Representations , year=

    LLMOPT: Learning to Define and Solve General Optimization Problems from Scratch , author=. The Thirteenth International Conference on Learning Representations , year=

  47. [55]

    Laurent Perron and Fr. The. 29th International Conference on Principles and Practice of Constraint Programming,

  48. [56]

    2011 , issn =

    An effective genetic algorithm for the flexible job-shop scheduling problem , journal =. 2011 , issn =

    2011

  49. [57]

    2001 , issn =

    An empirical investigation of meta-heuristic and heuristic algorithms for a 2D packing problem , journal =. 2001 , issn =

    2001

  50. [58]

    1993 , note =

    Benchmarks for basic scheduling problems , journal =. 1993 , note =

    1993

  51. [59]

    Chu, P. C. and Beasley, J. E. , title =. Journal of Heuristics , volume =

  52. [60]

    Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence , url=

    The Network Data Repository with Interactive Graph Analytics and Visualization , author=. Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence , url=

  53. [61]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Large Language Models as End-to-end Combinatorial Optimization Solvers , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

Pith tools

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