pith. sign in

arxiv: 2605.19119 · v1 · pith:OZAYUF5Hnew · submitted 2026-05-18 · 💻 cs.NE · cs.AI· cs.LG

GOAL: Graph-based Objective-Aligned Diffusion Solvers for Dynamic Multi-Objective Optimization

Xingyu Li This is my paper

Pith reviewed 2026-05-20 07:11 UTC · model grok-4.3

classification 💻 cs.NE cs.AIcs.LG
keywords diffusion modelsgraph neural networksmulti-objective optimizationscheduling problemscombinatorial optimizationjob shop schedulingfeasibilityconditioned generation
0
0 comments X

The pith

A conditioned diffusion model on heterogeneous graphs solves multi-objective scheduling problems with full feasibility and near-zero error across different constraint structures.

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

The paper proposes GOAL as a way to handle dynamic multi-objective optimization by turning solution generation into a diffusion process conditioned on specified objectives. It encodes problems as heterogeneous graphs where different edge types represent distinct constraint classes, allowing the graph neural network to pass messages selectively according to each constraint's structure. This setup is applied to three scheduling problems—flow shop, job shop, and flexible job shop—without any architecture changes, and the model produces feasible solutions while matching multiple objectives closely. A sympathetic reader would care because existing neural solvers are mostly stuck on single-objective static cases, while this one aims to deliver practical, controllable results on realistic combinatorial tasks.

Core claim

GOAL is a conditioned diffusion solver over relational graph representations that enables controllable decision generations by conditioning on human-specified objectives. It introduces a heterogeneous graph encoding in which distinct edge types, corresponding to different classes of constraints, define the message passing structure of the graph neural network, allowing information to propagate selectively. The approach generalizes across the Flow Shop Problem, Job Shop Scheduling Problem, and Flexible Job Shop Scheduling Problem without architectural modification. On these benchmarks for problem sizes up to 20 jobs and 60 operations, it reaches 100% solution feasibility and MAPE below 0.20%

What carries the argument

The heterogeneous graph encoding with distinct edge types for different constraint classes, which structures the message passing in the graph neural network to support selective information flow and objective-conditioned diffusion sampling.

If this is right

  • Solutions can be generated controllably by specifying different objective priorities at inference time rather than retraining the model.
  • The same trained model works on FSP, JSP, and FJSP despite their differing constraint structures.
  • Inference runs up to 25 times faster than NSGA-II or MOEA/D while producing higher-quality solutions.
  • All generated solutions satisfy every constraint by construction, eliminating post-processing repair steps.

Where Pith is reading between the lines

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

  • The conditioning mechanism could support interactive optimization where a user adjusts objectives on the fly during a production run.
  • Extending the edge-type ontology to include stochastic or time-varying constraints might allow handling of real-world uncertainty without redesigning the diffusion process.
  • Because the graph structure is the only place problem-specific information enters, similar encodings could transfer the method to other domains such as vehicle routing or resource allocation.

Load-bearing premise

The heterogeneous graph encoding with distinct edge types for different constraint classes enables selective message passing that supports generalization across structurally distinct problem types without any architectural modification.

What would settle it

Applying the unchanged GOAL model to a scheduling problem with an additional constraint class not seen in FSP, JSP, or FJSP and observing either feasibility below 100% or MAPE above 0.20% on the objectives.

Figures

Figures reproduced from arXiv: 2605.19119 by Xingyu Li.

Figure 1
Figure 1. Figure 1: Mean time to reach a qualified solution (milliseconds) across JSP instances of job size. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Time-to-ϵ across problem types and in￾stances. Error bars denote std over 100 instances. Tab. 2 reports Cmax MAPE, R MAPE, duplication rate, and feasibility across JSP, FSP, and FJSP for three instance sizes. GOAL achieves 100% feasibility with zero duplicate actions across all settings. For JSP and FSP, both MAPEs remain below 1.6% at all scales, while FJSP yields higher errors (up to 6.97% Cmax MAPE and … view at source ↗
Figure 4
Figure 4. Figure 4: Time to qualified decision grouped by machine counts [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
read the original abstract

Existing neural combinatorial optimization solvers frame solution search as imitation of optimal decisions, inherently limiting their utility to single-objective minimization and static constraints. We propose GOAL, a conditioned diffusion solver over relational graph representations that enables controllable decision generations by conditioning on human-specified objectives. We introduce a heterogeneous graph encoding in which distinct edge types, corresponding to different classes of constraints, define the message passing structure of the graph neural network, which allows information to propagate selectively according to the ontology of each constraint. GOAL is instantiated and evaluated on three canonical scheduling benchmarks of various constraint complexity: the Flow Shop Problem (FSP), the Job Shop Scheduling Problem (JSP), and the Flexible Job Shop Scheduling Problem (FJSP). Generalization is demonstrated across structurally distinct constraint regimes and problem types without architectural modification. On all three benchmarks, GOAL achieves 100% solution feasibility and near-zero MAPE (below 0.20%) on multiple objectives for problem sizes up to 20 jobs and 60 operations, outperforming NSGA-II and MOEA/D in both solution quality and inference speed by up to 25x.

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

2 major / 2 minor

Summary. The manuscript proposes GOAL, a conditioned diffusion solver that operates over heterogeneous relational graph representations of scheduling problems. Distinct edge types corresponding to different constraint classes structure the message passing in the underlying GNN, allowing the model to be conditioned on human-specified objectives for controllable multi-objective solution generation. The approach is evaluated on FSP, JSP, and FJSP benchmarks and claims to achieve 100% solution feasibility together with MAPE below 0.20% on multiple objectives for instances up to 20 jobs and 60 operations, while generalizing across these structurally distinct problem types without architectural modification and delivering up to 25x inference speedup relative to NSGA-II and MOEA/D.

Significance. If the reported performance and generalization results hold under rigorous controls, the work would represent a meaningful advance in neural combinatorial optimization by extending diffusion-based solvers to multi-objective, constraint-heterogeneous settings. The use of ontology-aware heterogeneous edges for selective message passing offers a concrete mechanism for incorporating problem structure that could improve feasibility and transfer across scheduling variants.

major comments (2)
  1. [Heterogeneous Graph Encoding] Heterogeneous Graph Encoding section: The central claim that distinct edge types for different constraint classes enable selective message passing that supports generalization across FSP, JSP, and FJSP without architectural changes is load-bearing for the reported 100% feasibility and cross-problem results. No ablation is presented that collapses the edge types into a single homogeneous relation while holding GNN architecture, diffusion conditioning, and training data fixed; without this comparison it remains unclear whether the performance arises from the claimed ontology-aware propagation or from overall model capacity.
  2. [Experimental results] Experimental results section: The manuscript reports 100% feasibility and MAPE below 0.20% together with 25x speedups, yet provides insufficient detail on the precise definitions and measurement procedures for feasibility and MAPE, the implementation of the NSGA-II and MOEA/D baselines, the number of independent runs, and any statistical testing. These omissions prevent verification that the quantitative claims are robustly supported by the data.
minor comments (2)
  1. [Abstract] The abstract would be strengthened by a single sentence summarizing the diffusion training objective or sampling procedure.
  2. [Method] Notation for the conditioning vector and edge-type embeddings should be introduced once and used consistently in all equations and figures.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and indicate the revisions that will be incorporated to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Heterogeneous Graph Encoding] Heterogeneous Graph Encoding section: The central claim that distinct edge types for different constraint classes enable selective message passing that supports generalization across FSP, JSP, and FJSP without architectural changes is load-bearing for the reported 100% feasibility and cross-problem results. No ablation is presented that collapses the edge types into a single homogeneous relation while holding GNN architecture, diffusion conditioning, and training data fixed; without this comparison it remains unclear whether the performance arises from the claimed ontology-aware propagation or from overall model capacity.

    Authors: We agree that an explicit ablation isolating the contribution of heterogeneous edge types would strengthen the central claim. In the revised manuscript we will add this ablation: all edge types will be collapsed into a single homogeneous relation while the GNN architecture, diffusion conditioning, and training data remain fixed. We will report the resulting feasibility, MAPE, and cross-problem generalization metrics to quantify the benefit of ontology-aware message passing. revision: yes

  2. Referee: [Experimental results] Experimental results section: The manuscript reports 100% feasibility and MAPE below 0.20% together with 25x speedups, yet provides insufficient detail on the precise definitions and measurement procedures for feasibility and MAPE, the implementation of the NSGA-II and MOEA/D baselines, the number of independent runs, and any statistical testing. These omissions prevent verification that the quantitative claims are robustly supported by the data.

    Authors: We acknowledge that additional experimental details are required for reproducibility and verification. In the revised manuscript we will expand the Experimental results section to specify: (i) exact definitions and procedures for feasibility (constraint satisfaction checks) and MAPE (mean absolute percentage error on each objective), (ii) full implementation details and hyper-parameters for the NSGA-II and MOEA/D baselines, (iii) the number of independent runs (five runs with distinct random seeds), and (iv) statistical significance tests (Wilcoxon rank-sum tests with p-values). These additions will substantiate the reported performance. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture and empirical claims are self-contained

full rationale

The paper introduces GOAL as a new conditioned diffusion solver using heterogeneous graph encodings for multi-objective scheduling. The abstract and described method present the heterogeneous edge types and selective message passing as an architectural choice enabling generalization across FSP/JSP/FJSP without modification, with performance (100% feasibility, <0.20% MAPE, 25x speedup) reported from benchmark evaluations rather than any derivation that reduces to fitted inputs or self-citations by construction. No equations, uniqueness theorems, or ansatzes are shown that equate the claimed results to prior fitted quantities or self-referential definitions. The central claims rest on empirical outcomes and the proposed ontology-aware GNN structure, which does not reduce to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated assumption that the described graph encoding and conditioning mechanism suffice for the reported generalization.

pith-pipeline@v0.9.0 · 5723 in / 1063 out tokens · 53899 ms · 2026-05-20T07:11:15.128829+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

84 extracted references · 84 canonical work pages · 4 internal anchors

  1. [1]

    Embodied intelligence via learning and evolution

    David Ha and Yujin Tang. Embodied intelligence via learning and evolution. InAdvances in Neural Information Processing Systems, 2023

    work page 2023

  2. [2]

    A generalist agent

    Scott Reed et al. A generalist agent. InTransactions on Machine Learning Research, 2022

    work page 2022

  3. [3]

    Champion-level drone racing using deep reinforcement learning

    Elia Kaufmann et al. Champion-level drone racing using deep reinforcement learning. 2023. 10

    work page 2023

  4. [4]

    Learning to walk in minutes using massively parallel deep reinforcement learning

    Nikita Rudin, David Hoeller, Philipp Reist, and Marco Hutter. Learning to walk in minutes using massively parallel deep reinforcement learning. InConference on Robot Learning, 2022

    work page 2022

  5. [5]

    Amazon. Amazon launches a new ai foundation model to power its robotic fleet and deploys its 1 millionth robot.https://www.aboutamazon.com/news/operations/ amazon-million-robots-ai-foundation-model, 2024. Accessed: 2025

    work page 2024

  6. [6]

    Industry 4.0.Business and Information Systems Engineering, 6(4):239–242, 2014

    Heiner Lasi, Peter Fettke, Hans-Georg Kemper, Thomas Feld, and Michael Hoffmann. Industry 4.0.Business and Information Systems Engineering, 6(4):239–242, 2014

    work page 2014

  7. [7]

    Malte Brettel, Niklas Friederichsen, Michael Keller, and Marius Rosenberg. How virtual- ization, decentralization and network building change the manufacturing landscape.Interna- tional Journal of Mechanical, Aerospace, Industrial and Mechatronics Engineering, 8(1):37– 44, 2014

    work page 2014

  8. [8]

    The surprising effectiveness of PPO in cooperative multi-agent games

    Chao Yu, Akash Velu, Eugene Vinitsky, Jiaxuan Gao, Yu Wang, Alexandre Bayen, and Yi Wu. The surprising effectiveness of PPO in cooperative multi-agent games. InAdvances in Neural Information Processing Systems, 2022

    work page 2022

  9. [9]

    Multi-agent deep reinforcement learning: A survey.Arti- ficial Intelligence Review, 55:895–943, 2022

    Sven Gronauer and Klaus Diepold. Multi-agent deep reinforcement learning: A survey.Arti- ficial Intelligence Review, 55:895–943, 2022

    work page 2022

  10. [10]

    Dynamic task allocation for multi-robot systems using graph neural networks

    Shengchao Liu et al. Dynamic task allocation for multi-robot systems using graph neural networks. InAdvances in Neural Information Processing Systems, 2023

    work page 2023

  11. [11]

    ScheduleNet: Learn to solve multi-agent scheduling problems with reinforcement learning

    Junyoung Park, Sanjar Bakhtiyar, and Jinkyoo Park. ScheduleNet: Learn to solve multi-agent scheduling problems with reinforcement learning. InarXiv preprint arXiv:2106.03051, 2021

    work page arXiv 2021

  12. [12]

    Self-labeling the job shop scheduling problem.Advances in Neural Information Processing Systems, 37:105528–105551, 2024

    Andrea Corsini, Angelo Porrello, Simone Calderara, and Mauro Dell’Amico. Self-labeling the job shop scheduling problem.Advances in Neural Information Processing Systems, 37:105528–105551, 2024

    work page 2024

  13. [13]

    Machine behaviour.Nature, 568(7753):477–486, 2019

    Iyad Rahwan, Manuel Cebrian, Nick Obradovich, Josh Bongard, Daniel C Dennett, Joanna Doyle, Simon Driessche, Jeffrey Dyer, Ziv Epstein, Eduardo Graells-Garrido, et al. Machine behaviour.Nature, 568(7753):477–486, 2019

    work page 2019

  14. [14]

    Coop- erative inverse reinforcement learning

    Dylan Hadfield-Menell, Smitha Milli, Pieter Abbeel, Stuart Russell, and Anca Dragan. Coop- erative inverse reinforcement learning. InAdvances in Neural Information Processing Systems, volume 29, pages 3909–3917, 2016

    work page 2016

  15. [15]

    Epureanu, Jian Cao, S

    Xingyu Li, Aydin Nassehi, Bogdan I. Epureanu, Jian Cao, S. Jack Hu, Lihui Wang, and Robert X. Gao. Generative manufacturing.Journal of Manufacturing Systems, 2026. Ac- cepted for publication, to appear

    work page 2026

  16. [16]

    Evolutionary dynamic multi- objective optimisation: A survey.ACM Computing Surveys, 55(4):1–47, 2022

    Shouyong Jiang, Juan Zou, Shengxiang Yang, and Xin Yao. Evolutionary dynamic multi- objective optimisation: A survey.ACM Computing Surveys, 55(4):1–47, 2022

    work page 2022

  17. [17]

    A fast and elitist mul- tiobjective genetic algorithm: NSGA-II.IEEE Transactions on Evolutionary Computation, 6(2):182–197, 2002

    Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and T Meyarivan. A fast and elitist mul- tiobjective genetic algorithm: NSGA-II.IEEE Transactions on Evolutionary Computation, 6(2):182–197, 2002

    work page 2002

  18. [18]

    Multiobjective optimization problems with complicated pareto sets, MOEA/D and NSGA-II.IEEE Transactions on Evolutionary Computation, 13(2):284–302, 2009

    Hui Li and Qingfu Zhang. Multiobjective optimization problems with complicated pareto sets, MOEA/D and NSGA-II.IEEE Transactions on Evolutionary Computation, 13(2):284–302, 2009

    work page 2009

  19. [19]

    Springer, 5th edition, 2016

    Michael L Pinedo.Scheduling: Theory, Algorithms, and Systems. Springer, 5th edition, 2016

    work page 2016

  20. [20]

    Evolutionary optimization in dynamic environments.Kluwer Academic Pub- lishers, 2001

    Jürgen Branke. Evolutionary optimization in dynamic environments.Kluwer Academic Pub- lishers, 2001

    work page 2001

  21. [21]

    Routing and scheduling in a flexible job shop by tabu search.Annals of Operations Research, 41(3):157–183, 1993

    Paolo Brandimarte. Routing and scheduling in a flexible job shop by tabu search.Annals of Operations Research, 41(3):157–183, 1993. 11

    work page 1993

  22. [22]

    Lee, Jimmy H.M

    Xinyi Hu, Jasper C.H. Lee, Jimmy H.M. Lee, and Peter J. Stuckey. Multi-stage pre- dict+optimize for (mixed integer) linear programs. InAdvances in Neural Information Pro- cessing Systems, volume 37, 2024

    work page 2024

  23. [23]

    Veronique Sels, Nele Gheysen, and Mario Vanhoucke. A comparison of priority rules for the job shop scheduling problem under different flow time-and tardiness-related objective func- tions.International Journal of Production Research, 50(15):4255–4270, 2012

    work page 2012

  24. [24]

    Review of job shop scheduling research and its new perspectives under industry 4.0.Journal of intelligent manu- facturing, 30(4):1809–1830, 2019

    Jian Zhang, Guofu Ding, Yisheng Zou, Shengfeng Qin, and Jianlin Fu. Review of job shop scheduling research and its new perspectives under industry 4.0.Journal of intelligent manu- facturing, 30(4):1809–1830, 2019

    work page 2019

  25. [25]

    An iterative improvement approach for the nonpreemptive open shop scheduling problem.European Journal of Operational Research, 111(3):509–517, 1998

    Ching-Fang Liaw. An iterative improvement approach for the nonpreemptive open shop scheduling problem.European Journal of Operational Research, 111(3):509–517, 1998

    work page 1998

  26. [26]

    MOEA/D: A multiobjective evolutionary algorithm based on de- composition.IEEE Transactions on Evolutionary Computation, 11(6):712–731, 2007

    Qingfu Zhang and Hui Li. MOEA/D: A multiobjective evolutionary algorithm based on de- composition.IEEE Transactions on Evolutionary Computation, 11(6):712–731, 2007

    work page 2007

  27. [27]

    John Wiley & Sons, 2009

    El-Ghazali Talbi.Metaheuristics: From Design to Implementation. John Wiley & Sons, 2009

    work page 2009

  28. [28]

    Many-objective optimization: An engineering design perspective

    Peter J Fleming, Robin C Purshouse, and Robert J Lygoe. Many-objective optimization: An engineering design perspective. InInternational Conference on Evolutionary Multi-Criterion Optimization, pages 14–32. Springer, 2005

    work page 2005

  29. [29]

    Multi-task learning as multi-objective optimization

    Ozan Sener and Vladlen Koltun. Multi-task learning as multi-objective optimization. InAd- vances in Neural Information Processing Systems, volume 31, 2018

    work page 2018

  30. [30]

    Pareto multi-task learn- ing

    Xi Lin, Hui-Ling Zhen, Zhenhua Li, Qingfu Zhang, and Sam Kwong. Pareto multi-task learn- ing. InAdvances in Neural Information Processing Systems, volume 32, 2019

    work page 2019

  31. [31]

    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

    work page 2021

  32. [32]

    Learning to solve combinatorial optimization problems on real-world graphs in linear time

    Dražen Vu ˇckovi´c et al. Learning to solve combinatorial optimization problems on real-world graphs in linear time. InAdvances in Neural Information Processing Systems, 2022

    work page 2022

  33. [33]

    Attention, learn to solve routing problems! InInternational Conference on Learning Representations, 2019

    Wouter Kool, Herke van Hoof, and Max Welling. Attention, learn to solve routing problems! InInternational Conference on Learning Representations, 2019

    work page 2019

  34. [34]

    Pointer networks

    Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. Pointer networks. InAdvances in Neural Information Processing Systems, volume 28, 2015

    work page 2015

  35. [35]

    Reinforce- ment learning for solving the vehicle routing problem

    Mohammadreza Nazari, Afshin Oroojlooy, Lawrence Snyder, and Martin Takác. Reinforce- ment learning for solving the vehicle routing problem. InAdvances in Neural Information Processing Systems, volume 31, 2018

    work page 2018

  36. [36]

    arXiv preprint arXiv:1906.01227 , year=

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

    work page arXiv 1906

  37. [37]

    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

    work page 2023

  38. [38]

    Expensive multi-objective bayesian optimization based on diffusion models

    Bingdong Li, Zixiang Di, Yongfan Lu, Hong Qian, Feng Wang, Peng Yang, Ke Tang, and Aimin Zhou. Expensive multi-objective bayesian optimization based on diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 27063–27071, 2025

    work page 2025

  39. [39]

    A new prediction strategy for dynamic multi-objective optimization using diffusion model.IEEE Transactions on Evolutionary Com- putation, 2025

    Feng Wang, Jinsong Xie, Aimin Zhou, and Ke Tang. A new prediction strategy for dynamic multi-objective optimization using diffusion model.IEEE Transactions on Evolutionary Com- putation, 2025. 12

    work page 2025

  40. [40]

    Diffusion learning-guided evolution for large- scale dynamic multi-objective optimization.IEEE Transactions on Evolutionary Computation, 2026

    Chenyang Li, Gary G Yen, and Zhenan He. Diffusion learning-guided evolution for large- scale dynamic multi-objective optimization.IEEE Transactions on Evolutionary Computation, 2026

    work page 2026

  41. [41]

    Jan Karel Lenstra and Alexander H. G. Rinnooy Kan.Complexity of Scheduling under Prece- dence Constraints, volume 26. INFORMS, 1977

    work page 1977

  42. [42]

    The complexity of flowshop and jobshop scheduling.Mathematics of Operations Research, 1(2):117–129, 1976

    Michael R Garey, David S Johnson, and Ravi Sethi. The complexity of flowshop and jobshop scheduling.Mathematics of Operations Research, 1(2):117–129, 1976

    work page 1976

  43. [43]

    Springer, Berlin, Heidelberg, 2nd edition, 2005

    Matthias Ehrgott.Multicriteria Optimization. Springer, Berlin, Heidelberg, 2nd edition, 2005

    work page 2005

  44. [44]

    The multiobjective traveling salesman problem: A survey and a new approach.Advances in Multi-Objective Nature Inspired Computing, pages 119–141, 2010

    Thibaut Lust and Jacques Teghem. The multiobjective traveling salesman problem: A survey and a new approach.Advances in Multi-Objective Nature Inspired Computing, pages 119–141, 2010

    work page 2010

  45. [45]

    The magical number seven, plus or minus two: Some limits on our capacity for processing information.Psychological Review, 63(2):81–97, 1956

    George A Miller. The magical number seven, plus or minus two: Some limits on our capacity for processing information.Psychological Review, 63(2):81–97, 1956

    work page 1956

  46. [46]

    SMAC3: A versatile bayesian optimization pack- age for hyperparameter optimization

    Marius Lindauer, Katharina Eggensperger, Matthias Feurer, André Biedenkapp, Difan Deng, Carolin Sass, Ed Bergman, and Frank Hutter. SMAC3: A versatile bayesian optimization pack- age for hyperparameter optimization. InJournal of Machine Learning Research, volume 23, pages 1–9, 2022

    work page 2022

  47. [47]

    The use of reference objectives in multiobjective optimization.Multiple Criteria Decision Making Theory and Application, pages 468–486, 1980

    Andrzej P Wierzbicki. The use of reference objectives in multiobjective optimization.Multiple Criteria Decision Making Theory and Application, pages 468–486, 1980

    work page 1980

  48. [48]

    Foundation models for decision making: Problems, methods, and opportunities, 2023

    Sherry Yang, Ofir Nachum, Yilun Du, Jason Wei, Pieter Abbeel, and Dale Schuurmans. Foun- dation models for decision making: Problems, methods, and opportunities. InarXiv preprint arXiv:2303.04129, 2023

    work page arXiv 2023

  49. [49]

    Llms can schedule.arXiv preprint arXiv:2408.06993, 2024

    Henrik Abgaryan, Ararat Harutyunyan, and Tristan Cazenave. Llms can schedule.arXiv preprint arXiv:2408.06993, 2024

    work page arXiv 2024

  50. [50]

    Automated scheduling heuristic gen- eration and evaluation via large language model.IEEE Transactions on Evolutionary Compu- tation, 2026

    Fei Yu, Liang Gao, Xinyu Li, Chao Lu, and Qihao Liu. Automated scheduling heuristic gen- eration and evaluation via large language model.IEEE Transactions on Evolutionary Compu- tation, 2026

    work page 2026

  51. [51]

    Dis- covering heuristics with large language models (llms) for mixed-integer programs: Single- machine scheduling.Computers & Operations Research, page 107325, 2025

    ˙Ibrahim O˘guz Çetinkaya, ˙I Esra Büyüktahtakın, Parshin Shojaee, and Chandan K Reddy. Dis- covering heuristics with large language models (llms) for mixed-integer programs: Single- machine scheduling.Computers & Operations Research, page 107325, 2025

    work page 2025

  52. [52]

    Multi-agent large language models as evolutionary optimizers for scheduling optimization.Computers & Industrial Engineering, 206:111197, 2025

    Yidan Wang, Jiayin Wang, and Zhiwei Chu. Multi-agent large language models as evolutionary optimizers for scheduling optimization.Computers & Industrial Engineering, 206:111197, 2025

    work page 2025

  53. [53]

    Language models are few-shot learn- ers

    Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agar- wal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Lit...

    work page 1901

  54. [54]

    LLaMA: Open and Efficient Foundation Language Models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Tim- othée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. LLaMA: Open and efficient foundation language models. InarXiv preprint arXiv:2302.13971, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  55. [55]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InAdvances in Neural Infor- mation Processing Systems, volume 30, 2017. 13

    work page 2017

  56. [56]

    Generative manu- facturing systems.SSRN Electron

    Xingyu Li, A Nassehi, H Yang, F Tao, J Sutherland, L Wang, and R Gao. Generative manu- facturing systems.SSRN Electron. J, 2025

    work page 2025

  57. [57]

    Efficient active search for combinato- rial optimization problems

    André Hottung, Yeong-Dae Kwon, and Kevin Murphy. Efficient active search for combinato- rial optimization problems. InInternational Conference on Learning Representations, 2022

    work page 2022

  58. [58]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020

    work page 2020

  59. [59]

    Struc- tured denoising diffusion models in discrete state-spaces

    Jacob Austin, Daniel D Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg. Struc- tured denoising diffusion models in discrete state-spaces. InAdvances in Neural Information Processing Systems, volume 34, 2021

    work page 2021

  60. [60]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

    work page 2021

  61. [61]

    Diffusion models for black-box optimization

    Siddarth Krishnamoorthy, Satvik Mehul Mashkaria, and Aditya Grover. Diffusion models for black-box optimization. InInternational Conference on Machine Learning, pages 17842– 17857. PMLR, 2023

    work page 2023

  62. [62]

    Diffsg: A generative solver for network optimization with diffusion model.arXiv preprint arXiv:2408.06701, 2024

    Ruihuai Liang, Bo Yang, Zhiwen Yu, Bin Guo, Xuelin Cao, Mérouane Debbah, H Vincent Poor, and Chau Yuen. Diffsg: A generative solver for network optimization with diffusion model.arXiv preprint arXiv:2408.06701, 2024

    work page arXiv 2024

  63. [63]

    Diffusion model for data-driven black-box optimization.arXiv preprint arXiv:2403.13219, 2024

    Zihao Li, Hui Yuan, Kaixuan Huang, Chengzhuo Ni, Yinyu Ye, Minshuo Chen, and Mengdi Wang. Diffusion model for data-driven black-box optimization.arXiv preprint arXiv:2403.13219, 2024

    work page arXiv 2024

  64. [64]

    Diff-bbo: Diffusion- based inverse modeling for black-box optimization.arXiv preprint arXiv:2407.00610, 2024

    Dongxia Wu, Nikki Lijing Kuang, Ruijia Niu, Yi-An Ma, and Rose Yu. Diff-bbo: Diffusion- based inverse modeling for black-box optimization.arXiv preprint arXiv:2407.00610, 2024

    work page arXiv 2024

  65. [65]

    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 optimiza- tion.Advances in Neural Information Processing Systems, 37:30179–30206, 2024

    work page 2024

  66. [66]

    A large manu- facturing decision model for human-centric decision-making.CIRP Annals, 2025

    Xingyu Li, Aydin Nassehi, S Jack Hu, Byung Gun Joung, and Robert X Gao. A large manu- facturing decision model for human-centric decision-making.CIRP Annals, 2025

    work page 2025

  67. [67]

    Comboptnet: Fit the right np-hard problem by learning integer programming constraints

    Anselm Paulus, Michal Rolínek, Vít Musil, Brandon Amos, and Georg Martius. Comboptnet: Fit the right np-hard problem by learning integer programming constraints. InInternational Conference on Machine Learning, pages 8443–8453. PMLR, 2021

    work page 2021

  68. [68]

    DC3: A learning method for optimization with hard constraints

    Priya L Donti, David Rolnick, and J Zico Kolter. DC3: A learning method for optimization with hard constraints. InInternational Conference on Learning Representations, 2021

    work page 2021

  69. [69]

    Learning to dispatch for job shop scheduling via deep reinforcement learning

    Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. Learning to dispatch for job shop scheduling via deep reinforcement learning. InAdvances in Neural Information Processing Systems, 2020

    work page 2020

  70. [70]

    Learning to delegate for large-scale vehicle routing

    Sirui Li, Zhongxia Yan, and Cathy Wu. Learning to delegate for large-scale vehicle routing. InAdvances in Neural Information Processing Systems, volume 34, 2021

    work page 2021

  71. [71]

    Learning the pareto front with hypernetworks

    Aviv Navon, Aviv Shamsian, Ethan Fetaya, and Gal Chechik. Learning the pareto front with hypernetworks. InInternational Conference on Learning Representations, 2021

    work page 2021

  72. [72]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  73. [73]

    Multilayer feedforward networks are universal approximators.Neural Networks, 2(5):359–366, 1989

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators.Neural Networks, 2(5):359–366, 1989

    work page 1989

  74. [74]

    Modeling relational data with graph convolutional networks

    Michael Schlichtkrull, Thomas N Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional networks. InEuropean Se- mantic Web Conference, pages 593–607, 2018. 14

    work page 2018

  75. [75]

    Relational inductive biases, deep learning, and graph networks

    Peter W. Battaglia et al. Relational inductive biases, deep learning, and graph networks.arXiv preprint arXiv:1806.01261, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  76. [76]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InAdvances in Neural Information Processing Systems, 2017

    work page 2017

  77. [77]

    Diffusion models beat GANs on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image synthesis. In Advances in Neural Information Processing Systems, volume 34, pages 8780–8794, 2021

    work page 2021

  78. [78]

    FiLM: Visual reasoning with a general conditioning layer

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron Courville. FiLM: Visual reasoning with a general conditioning layer. InProceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018

    work page 2018

  79. [79]

    Gaussian Error Linear Units (GELUs)

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (GELUs).arXiv preprint arXiv:1606.08415, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  80. [80]

    Schütt, Andreas Mayr, Michael Gastegger, and Philipp Marquetand

    Justin Gilmer, Kristof T. Schütt, Andreas Mayr, Michael Gastegger, and Philipp Marquetand. Neural message passing for quantum chemistry. InInternational Conference on Machine Learning, 2017

    work page 2017

Showing first 80 references.