Pith. sign in

REVIEW 3 major objections 4 minor 61 references

GPLight+: A Genetic Programming Method for Learning Symmetric Traffic Signal Control Policy

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that imposing a permutation-symmetry constraint on the phase urgency function significantly improves GP-learned traffic signal control policies on most public benchmark scenarios.

desk verdict Solid, well-executed evolutionary TSC paper held back by a confounded design: the symmetry claim isn't isolated from search-space compression and the additive aggregator. read the letter →

arxiv 2508.16090 v1 pith:I4ZEF2Q7 submitted 2025-08-22 cs.LG cs.AI

Xiao-Cheng Liao , Yi Mei , Mengjie Zhang This is my paper
classification cs.LGcs.AI
keywords trafficsignalcontrolgeneticprogrammingsymmetryconstraintphaseurgencyfunctioninterpretablepolicyevolutionaryoptimizationCityFlow
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 aims to improve genetic-programming-based traffic signal control by enforcing a symmetry constraint on the urgency function that decides which signal phase gets the green light next. Existing GP methods feed the features of the two turn movements in a phase into the function in a fixed order, so swapping the order can change the computed urgency even though the phase is unchanged. The paper argues this order-dependence makes policies inconsistent and forces GP to search a larger, noisier space. The proposed method, GPLight+, evolves a single turn-movement urgency subtree and uses two structurally identical copies of it, one per movement, added together to form the phase urgency, making the function symmetric by construction. In the CityFlow simulator on six public real-world datasets, the symmetric representation gives significantly lower average travel time than GPLight on five of the six, with the largest gain on the Manhattan network, and also yields policies that generalize across scenarios and run on microcontroller-class hardware.

What carries the argument

The central object is the symmetric phase urgency function. A phase contains two unordered turn movements (TMs), and its urgency is the sum of two TM urgencies, where each TM urgency is computed by the same tree-based function over that movement's lane features (waiting vehicles and total vehicle count on the incoming and relevant outgoing lanes). The two subtrees are forced to be structurally identical (isomorphic copies), and a fixed lane-feature permutation template ensures the same TM always receives its features in the same order across all phases. This construction makes the phase urgency provably invariant to the order of the two TMs, a property the asymmetric GPLight function lacks.

What would settle it

Re-run both methods on the six benchmark datasets and check the Wilcoxon test: the paper's claim predicts GPLight+ wins significantly on at least 5 of 6. A flip in the majority—or a new dataset where asymmetric GPLight consistently beats GPLight+ in 30-run average travel time—would falsify the symmetry-benefit claim as stated.

Watch

Extended reading notes

Core claim

The central claim is that the phase urgency for a signal phase at an intersection should be computed by a symmetric function of the two turn movements it comprises, and that encoding this symmetry directly into the GP representation improves the learned control policy. Concretely, the phase urgency is defined as the sum of two identical subtree values, each subtree (the TM urgency function) reading only features of one turn movement, ordered by a fixed lane-feature permutation template. Because the two subtrees are structurally identical, the function automatically satisfies swap-invariance: exchanging the two movements leaves the phase urgency unchanged. The paper demonstrates, over 30 inde

Load-bearing premise

The phase urgency is modelled as a plain sum of two independent single-turn-movement urgencies; if the two movements' effects interact (for example, both feed the same downstream bottleneck), that additive structure caps how expressive the learned policy can be.

Editorial extensions

If this is right

  • If the results hold, traffic signal control policies can be evolved as compact arithmetic expressions and deployed on low-cost microcontrollers without quantization, with far fewer parameters and FLOPs than DRL policies.
  • The symmetric representation improves data efficiency in evolution: the paper shows better initialization and faster convergence in five of six scenarios, implying less simulation time to reach a given policy quality.
  • The learned policies remain human-understandable; an evolved TM urgency function such as max(W0 - C0, 0.9) can be inspected directly, showing that incoming-lane queue features dominate and that congested right-turn outgoing lanes reduce urgency.
  • Generalization improves: lower variance when transferring a policy trained on one Jinan dataset to another suggests the symmetry constraint yields more robust, less overfit policies than DRL baselines.
  • The largest gains appear in the large-scale Manhattan network, suggesting the symmetry benefit becomes more pronounced as the number of controlled intersections grows.

Reading between the lines

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

  • The shared-subtree, permutation-invariant scaffold is a general pattern: any GP-evolved decision heuristic whose inputs form an unordered set of sub-objects (scheduling jobs, routing nodes, candidate moves) could benefit from the same treatment, though the paper itself does not claim this.
  • The fixed additive aggregator is the paper's acknowledged weakest point; a natural testable extension is to evolve or co-adapt a symmetric aggregator (multiplication, max, average) while keeping the shared-subtree constraint, to see whether the symmetry benefit survives a more expressive combination rule.
  • Because the paper's benefit is largest on a many-intersection network, a follow-up experiment on a heavily congested network where the two movements discharge into a shared downstream bottleneck would stress the additive-independence assumption; if GPLight+ still wins there, the symmetry thesis is more robust than the simple additive interpretation alone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes GPLight+, a genetic programming method for traffic signal control. Unlike GPLight, which evolves a phase-urgency tree over the features of both turn movements of an 8-phase intersection, GPLight+ evolves a single TM-urgency subtree, duplicates it for the two TMs of a phase, and sums the two outputs to obtain the phase urgency. The shared-subtree construction guarantees that the phase-urgency function is invariant to the order of the two TMs. The method is evaluated in CityFlow on six public real-world datasets with 30 independent runs, reporting average travel time (ATT), average queue length (AQL), and network throughput (NT), together with convergence, generalization, and resource-consumption analyses. The paper claims significant ATT improvements over GPLight on 5 of 6 datasets, with the largest reduction on Manhattan, and that the evolved policies are human-understandable and deployable on microcontrollers.

Significance. If the causal claim about the symmetry constraint were established, this would be a valuable empirical result: a simple representational prior improves performance while producing smaller, interpretable policies. The paper includes a code link, multiple public benchmarks, 30-run statistics, and significance tests, which are strengths. The main contribution is a design choice rather than a theoretical result, and its significance depends on isolating the symmetry mechanism. As it stands, the experimental design does not support the title-level attribution to symmetry, so the result is promising but not fully established.

major comments (3)
  1. [§IV.E, Eq. (12), Fig. 7] The central claim that the symmetry constraint causes the improvement is not supported by the experimental design. GPLight+ differs from GPLight in three coupled ways: (i) the two TM subtrees are forced to be identical; (ii) the terminal set is halved from 16 to 8 features (Section III.D, Eq. (14)); and (iii) the phase urgency is restricted to the sum of two TM urgencies (Eq. (12)) instead of an arbitrary function over both TMs. Tables III and IV compare only the complete systems. The authors themselves identify the search-space compression as a 'potential benefit' in Section IV.E and the additive aggregator as a 'potential performance limitation' in Section V. Without an ablation that varies one mechanism at a time—e.g., a GPLight variant with the same 8-terminal TM-level representation but no subtree sharing, or GPLight+ with different symmetric aggregators—the observed gains cannot be
  2. [§III.B.2, Eq. (12)] The choice of the additive aggregator is not justified and is entangled with the symmetry claim. Eq. (3) only requires f_phase(x1,x2)=f_phase(x2,x1); sharing subtrees plus any symmetric aggregator satisfies this. Fixing the aggregator to addition restricts the class of representable phase urgencies, and the paper acknowledges this in Section V but provides no experiment on alternative aggregators. A comparison with a learned or evolved symmetric aggregator, or even with a max or product aggregator, would clarify whether the reported improvement is due to symmetry or to the additive form. Without this, the phrase 'represented as an aggregation of two shared subtrees' in the abstract describes an additional design choice rather than the symmetry contribution itself.
  3. [§IV.G, Fig. 10] The generalization claim ('satisfying the symmetry ... can effectively enhance its generalizability') is based on visual inspection of variances in Fig. 10, with no statistical test and no quantitative summary. The text says 'in a few scenarios' GPLight exhibits greater variance than GPLight+, which is too vague to support the claim. Report the distributions of transfer ATT differences and test differences in means or variances (e.g., paired tests across scenarios) before making this assertion. This is a secondary claim, but as written it overstates what the cross-validation demonstrates.
minor comments (4)
  1. [Tables III and IV footnote] The footnote says the symbols '+/-/≈' indicate whether a metric is 'significantly worse than, better than or statistically comparable to the proposed GPLight algorithm.' This appears to refer to GPLight+, and the direction of '+' and '-' relative to the proposed method should be stated unambiguously.
  2. [Throughout] Many mathematical symbols appear as missing glyphs in the rendered text, especially in Eqs. (1)-(3), Fig. 1, and the dataset names in Table IV (e.g., 'Hangzhou�'). Please fix the encoding so the phase names and formulas are readable.
  3. [§IV.E] The text states that GPLight+ 'obtains the smallest queue length in all scenarios.' This is contradicted by Table III for Jinan1, where PressLight has AQL 11.0108 vs GPLight+ 12.0057. Similarly, 'highest throughput in most scenarios' needs qualification: in Jinan1, GPLight+ NT 6124.7 is below MPLight (6146.1), FRAP (6147.3), and PressLight (6163.7).
  4. [§III.B.4 and §IV.G] Section III.B.4 refers to 'Fig. S7' for the feature permutation template; the main text should be self-contained or the supplementary material should be clearly part of the reviewed artifact. Also, 'preformed' in Section IV.G should be 'performed.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical performance comparison; the symmetry constraint is a design prior, not a derived or fitted quantity.

full rationale

The paper's central claim is comparative: GPLight+'s symmetric phase-urgency representation improves over GPLight on measured average travel time in CityFlow simulations. Nothing in the derivation defines the outcome in terms of the input. The symmetric representation is proposed (Eq. 3 and Fig. 7) and evolved by GP, then evaluated on real-world traffic datasets; the reported gains are measured, not entailed by construction. The paper does change several coupled design choices relative to GPLight—shared symmetric subtrees, a halved terminal set, and a fixed additive aggregator (Eq. 12)—and no ablation isolates symmetry alone. The authors themselves flag this in Section IV.E ('the introduction of the symmetry constraint, to some extent, restricts the search space of phase urgency functions') and Section V ('this work utilizes a simple addition as the aggregator for phase urgency function, which could lead to potential performance limitations'). This is an attribution/confound limitation, not circularity: no fitted parameter is renamed as a prediction, no equation reduces to its own input, and no load-bearing premise depends on a self-citation. The only self-citation, GPLight [18], is used as a baseline and is empirically re-evaluated in the same simulator, so it is independent support rather than circular justification. The limitation passages were reviewed and do not reveal any definitional equivalence.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical entities or external parameters beyond the design choices of the GP representation. The main design commitments are the additive aggregator, the symmetry constraint, and the feature ordering template. GP hyperparameters and ephemeral constants are standard to the evolutionary search.

free parameters (2)
  • GP hyperparameters = population size=100, generations=51, max depth=6, crossover=90%, mutation=10%
    Chosen by hand following prior work [18] and standard GP practice; not tuned to the specific datasets, but they affect the search trajectory and can influence relative performance.
  • Ephemeral constants in evolved trees = Example: alpha=0.9 in Eq. (16)
    Random constants are introduced during GP evolution and are effectively fitted by selection. They are part of the evolved policy, not external parameters of the proposed method.
assumptions (3)
  • domain assumption Phase urgency can be decomposed as a sum of independent turn-movement urgencies
    The aggregator is fixed to addition (Eq. 12 and Fig. 7). This limits expressiveness and is acknowledged as a potential limitation in Section V.
  • domain assumption The two turn movements in a phase are unordered and should be treated symmetrically
    Motivates the symmetry constraint. The paper argues that swapping the two TMs should not change urgency, which is a modeling choice about the problem structure.
  • domain assumption The lane feature permutation template ensures consistent feature ordering across phases
    Section III.B.4 and Fig. S7 define a fixed ordering of lanes and features. The effectiveness of the method depends on this template being semantically valid for all phases.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GPLight+: A Genetic Programming Method for Learning Symmetric Traffic Signal Control Policy." pith.science (2026). https://pith.science/paper/I4ZEF2Q7

@misc{pith2026250816090,
  author       = {Pith},
  title        = {Pith review of: GPLight+: A Genetic Programming Method for Learning Symmetric Traffic Signal Control Policy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I4ZEF2Q7}},
  note         = {Machine review of arXiv:2508.16090}
}
read the original abstract

Recently, learning-based approaches, have achieved significant success in automatically devising effective traffic signal control strategies. In particular, as a powerful evolutionary machine learning approach, Genetic Programming (GP) is utilized to evolve human-understandable phase urgency functions to measure the urgency of activating a green light for a specific phase. However, current GP-based methods are unable to treat the common traffic features of different traffic signal phases consistently. To address this issue, we propose to use a symmetric phase urgency function to calculate the phase urgency for a specific phase based on the current road conditions. This is represented as an aggregation of two shared subtrees, each representing the urgency of a turn movement in the phase. We then propose a GP method to evolve the symmetric phase urgency function. We evaluate our proposed method on the well-known cityflow traffic simulator, based on multiple public real-world datasets. The experimental results show that the proposed symmetric urgency function representation can significantly improve the performance of the learned traffic signal control policies over the traditional GP representation on a wide range of scenarios. Further analysis shows that the proposed method can evolve effective, human-understandable and easily deployable traffic signal control policies.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 51 canonical work pages

  1. [1]

    Combining traffic assignment and traffic signal control for online traffic flow optimization,

    X.-C. Liao, W.-J. Qiu, F.-F. Wei, and W.-N. Chen, “Combining traffic assignment and traffic signal control for online traffic flow optimization,” in Neural Information Processing , M. Tanveer, S. Agarwal, S. Ozawa, A. Ekbal, and A. Jatowt, Eds. Singapore: Springer Nature Singapore, 2023, pp. 150–163

    work page 2023

  2. [2]

    Lowrie, Roads, and T

    P. Lowrie, Roads, and T. A. of New South Wales. Traffic Control Section, SCATS, Sydney Co-Ordinated Adaptive Traffic System: A Traffic Responsive Method of Controlling Urban Traffic . Roads and Traffic Authority NSW, Traffic Control Section, 1990. [Online]. Available: https://books.google.co.nz/books?id=V4PTtgAACAAJ

    work page 1990

  3. [3]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,” nature, vol. 521, no. 7553, pp. 436–444, 2015

    2015

  4. [4]

    Sur- vey on the internet of vehicles: Network architectures and applications,

    B. Ji, X. Zhang, S. Mumtaz, C. Han, C. Li, H. Wen, and D. Wang, “Sur- vey on the internet of vehicles: Network architectures and applications,” IEEE Communications Standards Magazine , vol. 4, no. 1, pp. 34–41, 2020

    work page 2020

  5. [5]

    Toward a thousand lights: Decentralized deep reinforcement learning for large-scale traffic signal control,

    C. Chen, H. Wei, N. Xu, G. Zheng, M. Yang, Y . Xiong, K. Xu, and Z. Li, “Toward a thousand lights: Decentralized deep reinforcement learning for large-scale traffic signal control,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 34, no. 04, 2020, pp. 3414–3421

    work page 2020

  6. [6]

    Colight: Learning network-level cooperation for traffic signal control,

    H. Wei, N. Xu, H. Zhang, G. Zheng, X. Zang, C. Chen, W. Zhang, Y . Zhu, K. Xu, and Z. Li, “Colight: Learning network-level cooperation for traffic signal control,” in Proceedings of the 28th ACM International Conference on Information and Knowledge Management , 2019, pp. 1913–1922

    work page 2019

  7. [7]

    Presslight: Learning max pressure control to coordinate traffic signals in arterial network,

    H. Wei, C. Chen, G. Zheng, K. Wu, V . Gayah, K. Xu, and Z. Li, “Presslight: Learning max pressure control to coordinate traffic signals in arterial network,” in Proceedings of the 25th ACM SIGKDD inter- national conference on knowledge discovery & data mining , 2019, pp. 1290–1298

    work page 2019

  8. [8]

    Deep reinforcement learning for intelligent transportation systems: A survey,

    A. Haydari and Y . Yılmaz, “Deep reinforcement learning for intelligent transportation systems: A survey,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 1, pp. 11–32, 2020

    2020

Show all 61 references
  1. [9]

    Reward design with language models,

    M. Kwon, S. M. Xie, K. Bullard, and D. Sadigh, “Reward design with language models,” arXiv preprint arXiv:2303.00001 , 2023

    2023 arXiv

  2. [10]

    Inverse reward design,

    D. Hadfield-Menell, S. Milli, P. Abbeel, S. J. Russell, and A. Dragan, “Inverse reward design,” Advances in neural information processing systems, vol. 30, 2017

    2017

  3. [11]

    Evolution strategies as a scalable alternative to reinforcement learning,

    T. Salimans, J. Ho, X. Chen, S. Sidor, and I. Sutskever, “Evolution strategies as a scalable alternative to reinforcement learning,” arXiv preprint arXiv:1703.03864, 2017

    2017 arXiv

  4. [12]

    Explainable artificial intelligence (xai): What we know and what is left to attain trustworthy artificial intelligence,

    S. Ali, T. Abuhmed, S. El-Sappagh, K. Muhammad, J. M. Alonso- Moral, R. Confalonieri, R. Guidotti, J. Del Ser, N. D ´ıaz-Rodr´ıguez, and F. Herrera, “Explainable artificial intelligence (xai): What we know and what is left to attain trustworthy artificial intelligence,” Inform...

    2023

  5. [13]

    Interpretation of neural networks is fragile,

    A. Ghorbani, A. Abid, and J. Zou, “Interpretation of neural networks is fragile,” in Proceedings of the AAAI conference on artificial intelligence, vol. 33, no. 01, 2019, pp. 3681–3688

    2019

  6. [14]

    Extracting decision tree from trained deep reinforcement learning in traffic signal control,

    Y . Zhu, X. Yin, and C. Chen, “Extracting decision tree from trained deep reinforcement learning in traffic signal control,” IEEE Transactions on Computational Social Systems , vol. 10, no. 4, pp. 1997–2007, 2023

    1997

  7. [15]

    ’why did you do that?’ explainable intelligent robots,

    R. K.-M. Sheh, “’why did you do that?’ explainable intelligent robots,” in Workshops at the Thirty-First AAAI Conference on Artificial Intelli- gence, 2017, pp. 628–634

    2017

  8. [16]

    Explainable artifi- cial intelligence by genetic programming: A survey,

    Y . Mei, Q. Chen, A. Lensen, B. Xue, and M. Zhang, “Explainable artifi- cial intelligence by genetic programming: A survey,” IEEE Transactions on Evolutionary Computation , vol. 27, no. 3, pp. 621–641, 2023

    2023

  9. [17]

    Genetic programming as a means for programming comput- ers by natural selection,

    J. R. Koza, “Genetic programming as a means for programming comput- ers by natural selection,” Statistics and computing , vol. 4, pp. 87–112, 1994

    1994

  10. [18]

    Learning traffic signal control via genetic programming,

    X.-C. Liao, Y . Mei, and M. Zhang, “Learning traffic signal control via genetic programming,” in Proceedings of the Genetic and Evolutionary Computation Conference , ser. GECCO ’24. New York, NY , USA: Association for Computing Machinery, 2024, p. 924–932

    2024

  11. [19]

    Diagnosing reinforcement learning for traffic signal control,

    G. Zheng, X. Zang, N. Xu, H. Wei, Z. Yu, V . Gayah, K. Xu, and Z. Li, “Diagnosing reinforcement learning for traffic signal control,” arXiv preprint arXiv:1905.04716 , 2019

    1905 arXiv

  12. [20]

    Intellilight: A reinforcement learning approach for intelligent traffic light control,

    H. Wei, G. Zheng, H. Yao, and Z. Li, “Intellilight: A reinforcement learning approach for intelligent traffic light control,” in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , 2018, pp. 2496–2505

    2018

  13. [21]

    Learning phase competition for traffic signal control,

    G. Zheng, Y . Xiong, X. Zang, J. Feng, H. Wei, H. Zhang, Y . Li, K. Xu, and Z. Li, “Learning phase competition for traffic signal control,” in Proceedings of the 28th ACM international conference on information and knowledge management , 2019, pp. 1963–1972

    2019

  14. [22]

    Traffic signal settings,

    F. V . Webster, “Traffic signal settings,” Tech. Rep. 39, 1958

    1958

  15. [23]

    Efficient pressure: Improving efficiency for signalized intersections,

    Q. Wu, L. Zhang, J. Shen, L. L ¨u, B. Du, and J. Wu, “Efficient pressure: Improving efficiency for signalized intersections,” 2021. [Online]. Available: https://arxiv.org/abs/2112.02336

    2021 arXiv

  16. [24]

    Vissim: A microscopic simulation tool to evaluate actuated signal control including bus priority,

    M. Fellendorf, “Vissim: A microscopic simulation tool to evaluate actuated signal control including bus priority,” in 64th Institute of transportation engineers annual meeting , vol. 32. Springer, 1994, pp. 1–9

    1994

  17. [25]

    A real-time traffic signal control system: architecture, algorithms, and analysis,

    P. Mirchandani and L. Head, “A real-time traffic signal control system: architecture, algorithms, and analysis,” Transportation Research Part C: Emerging Technologies, vol. 9, no. 6, pp. 415–432, 2001

    2001

  18. [26]

    Traffic signal timing manual

    P. Koonce and L. Rodegerdts, “Traffic signal timing manual.” United States. Federal Highway Administration, Tech. Rep. FHW A-HOP-08- 024, 2008

    2008

  19. [27]

    The scoot on-line traffic signal optimisation technique,

    P. Hunt, D. Robertson, R. Bretherton, and M. C. Royle, “The scoot on-line traffic signal optimisation technique,” Traffic Engineering & Control, vol. 23, no. 4, 1982

    1982

  20. [28]

    Automatic design of optimal actuated traffic signal plans with active transit priority,

    M. Keblawi and T. Toledo, “Automatic design of optimal actuated traffic signal plans with active transit priority,”IEEE Transactions on Intelligent Transportation Systems, vol. 25, no. 6, pp. 5767–5779, 2024

    2024

  21. [29]

    Traffic signals,

    F. V . Webster and B. M. Cobbe, “Traffic signals,” Road Research Laboratory, Ministry of Transport, Tech. Rep. 56, 1966. [Online]. Available: https://nou.sinaldetransito.com.br/wp-content/uploads/2021/ 03/webster2.pdf

    1966

  22. [30]

    R. P. Roess, Traffic engineering. United states of Anerica, 2004

    2004

  23. [31]

    Signal multiobjective optimization for urban traffic network,

    X. Li and J.-Q. Sun, “Signal multiobjective optimization for urban traffic network,” IEEE Transactions on Intelligent Transportation Systems , vol. 19, no. 11, pp. 3529–3537, 2018

    2018

  24. [32]

    Novel traffic signal timing adjustment strategy based on genetic algorithm,

    H.-Y . Tung, W.-C. Ma, and T.-L. Yu, “Novel traffic signal timing adjustment strategy based on genetic algorithm,” in 2014 IEEE Congress on Evolutionary Computation (CEC) . IEEE, 2014, pp. 2353–2360

    2014

  25. [33]

    Multi-objective optimization of urban road intersection signal timing based on particle swarm optimization algorithm,

    H. Jia, Y . Lin, Q. Luo, Y . Li, and H. Miao, “Multi-objective optimization of urban road intersection signal timing based on particle swarm optimization algorithm,” Advances in Mechanical Engineering, vol. 11, no. 4, 2019. [Online]. Available: https: //doi.org/10.1177/1687814...

    2019 doi

  26. [34]

    Evolvable traffic signal control for intersection congestion alleviation with enhanced particle swarm optimisation,

    H. S. E. Chuo, M. K. Tan, A. C. H. Chong, R. K. Y . Chin, and K. T. K. Teo, “Evolvable traffic signal control for intersection congestion alleviation with enhanced particle swarm optimisation,” in 2017 IEEE 2nd International Conference on Automatic Control and Intelligent Syst...

    2017

  27. [35]

    A differential evolution algorithm-based traffic control model for signalized intersections,

    Z. Cakici and Y . S. Murat, “A differential evolution algorithm-based traffic control model for signalized intersections,” Advances in Civil Engineering, vol. 2019, pp. 1–16, 2019

    2019

  28. [36]

    A multiobjective bilevel programming model for environ- mentally friendly traffic signal timings,

    O. Baskan, “A multiobjective bilevel programming model for environ- mentally friendly traffic signal timings,” Advances in Civil Engineering , vol. 2019, pp. 1–13, 2019

    2019

  29. [37]

    Ant colony optimization approach for optimizing traffic signal timings,

    O. Baskan and S. Haldenbilen, “Ant colony optimization approach for optimizing traffic signal timings,” in Ant Colony Optimization , A. Ostfeld, Ed. Rijeka: IntechOpen, 2011, ch. 14. [Online]. Available: https://doi.org/10.5772/13665

    2011 doi

  30. [38]

    Traffic signal optimization using ant colony algorithm,

    D. Renfrew and X.-H. Yu, “Traffic signal optimization using ant colony algorithm,” in The 2012 International Joint Conference on Neural Networks (IJCNN). IEEE, 2012, pp. 1–7

    2012

  31. [39]

    The max-pressure controller for arbitrary networks of signalized intersections,

    P. Varaiya, “The max-pressure controller for arbitrary networks of signalized intersections,” in Advances in dynamic network modeling in complex transportation systems . Springer, 2013, pp. 27–66

    2013

  32. [40]

    Safelight: A rein- forcement learning method toward collision-free traffic signal control,

    W. Du, J. Ye, J. Gu, J. Li, H. Wei, and G. Wang, “Safelight: A rein- forcement learning method toward collision-free traffic signal control,” in Proceedings of the AAAI conference on artificial intelligence , vol. 37, no. 12, 2023, pp. 14 801–14 810

    2023

  33. [41]

    Reinforcement learning with function approximation for traffic signal control,

    L. Prashanth and S. Bhatnagar, “Reinforcement learning with function approximation for traffic signal control,” IEEE Transactions on Intelli- gent Transportation Systems, vol. 12, no. 2, pp. 412–421, 2010

    2010

  34. [42]

    Multiagent rein- forcement learning for urban traffic control using coordination graphs,

    L. Kuyer, S. Whiteson, B. Bakker, and N. Vlassis, “Multiagent rein- forcement learning for urban traffic control using coordination graphs,” in Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2008, Antwerp, Belgium, September 15-19, 2008, ...

    2008

  35. [43]

    Coordinated deep reinforcement learners for traffic light control,

    E. Van der Pol and F. A. Oliehoek, “Coordinated deep reinforcement learners for traffic light control,” Proceedings of learning, inference and control of multi-agent systems (at NIPS 2016) , vol. 8, pp. 21–38, 2016. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2024 15

    2016

  36. [44]

    Multi-agent reinforcement learning for traffic light control,

    M. A. Wiering et al. , “Multi-agent reinforcement learning for traffic light control,” in Machine Learning: Proceedings of the Seventeenth International Conference (ICML) , 2000, pp. 1151–1158

    2000

  37. [45]

    Learning heuristics with different rep- resentations for stochastic routing,

    Y .-H. Jia, Y . Mei, and M. Zhang, “Learning heuristics with different rep- resentations for stochastic routing,” IEEE Transactions on Cybernetics , vol. 53, no. 5, pp. 3205–3219, 2023

    2023

  38. [46]

    Survey on genetic programming and machine learning techniques for heuristic design in job shop scheduling,

    F. Zhang, Y . Mei, S. Nguyen, and M. Zhang, “Survey on genetic programming and machine learning techniques for heuristic design in job shop scheduling,” IEEE Transactions on Evolutionary Computation, vol. 28, no. 1, pp. 147–167, 2023

    2023

  39. [47]

    Zhang, S

    F. Zhang, S. Nguyen, Y . Mei, and M. Zhang, Genetic Programming for Production Scheduling: An Evolutionary Learning Approach , ser. Machine Learning: Foundations, Methodologies, and Applications. Sin- gapore: Springer, 2021

    2021

  40. [48]

    Genetic programming for production scheduling: a survey with a unified framework,

    S. Nguyen, Y . Mei, and M. Zhang, “Genetic programming for production scheduling: a survey with a unified framework,” Complex & Intelligent Systems, vol. 3, pp. 41–66, 2017

    2017

  41. [49]

    Niching genetic programming to learn actions for deep reinforcement learning in dynamic flexible scheduling,

    M. Xu, Y . Mei, F. Zhang, and M. Zhang, “Niching genetic programming to learn actions for deep reinforcement learning in dynamic flexible scheduling,” IEEE Transactions on Evolutionary Computation , 2024

    2024

  42. [50]

    Genetic programming and reinforcement learning on learning heuristics for dynamic scheduling: A preliminary comparison,

    ——, “Genetic programming and reinforcement learning on learning heuristics for dynamic scheduling: A preliminary comparison,” IEEE Computational Intelligence Magazine , vol. 19, no. 2, pp. 18–33, 2024

    2024

  43. [51]

    A genetic programming approach for the traffic signal control problem with epigenetic modifications,

    E. Ricalde and W. Banzhaf, “A genetic programming approach for the traffic signal control problem with epigenetic modifications,” in Genetic Programming, Springer. Cham: Springer International Publishing, 2016, pp. 133–148

    2016

  44. [52]

    Evolving adaptive traffic signal controllers for a real scenario using genetic programming with an epigenetic mechanism,

    ——, “Evolving adaptive traffic signal controllers for a real scenario using genetic programming with an epigenetic mechanism,” in2017 16th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2017, pp. 897–902

    2017

  45. [53]

    A survey and comparison of tree generation algorithms,

    S. Luke and L. Panait, “A survey and comparison of tree generation algorithms,” in Proceedings of the 3rd Annual Conference on Genetic and Evolutionary Computation , 2001, pp. 81–88

    2001

  46. [54]

    Expression might be enough: Representing pressure and demand for reinforcement learning based traffic signal control,

    L. Zhang, Q. Wu, J. Shen, L. L ¨u, B. Du, and J. Wu, “Expression might be enough: Representing pressure and demand for reinforcement learning based traffic signal control,” in International Conference on Machine Learning. PMLR, 2022, pp. 26 645–26 654

    2022

  47. [55]

    Towards scalable dynamic traffic assignment with streaming agents: A decentralized control approach using genetic programming,

    X.-C. Liao, W.-N. Chen, Y .-H. Jia, and W.-J. Qiu, “Towards scalable dynamic traffic assignment with streaming agents: A decentralized control approach using genetic programming,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 8, no. 1, pp. 942– 955, 2024

    2024

  48. [56]

    Learning to communicate among agents for large-scale dynamic path planning with genetic programming hyperheuristic,

    X.-C. Liao, X.-M. Hu, X.-L. Chen, Y . Mei, Y .-H. Jia, and W.-N. Chen, “Learning to communicate among agents for large-scale dynamic path planning with genetic programming hyperheuristic,” IEEE Transactions on Artificial Intelligence , vol. 6, no. 5, pp. 1269–1283, 2025

    2025

  49. [57]

    Cityflow: A multi-agent reinforcement learning environment for large scale city traffic scenario,

    H. Zhang, S. Feng, C. Liu, Y . Ding, Y . Zhu, Z. Zhou, W. Zhang, Y . Yu, H. Jin, and Z. Li, “Cityflow: A multi-agent reinforcement learning environment for large scale city traffic scenario,” in The world wide web conference, 2019, pp. 3620–3624

    2019

  50. [58]

    Tinylight: Adaptive traffic signal control on devices with extremely limited resources,

    D. Xing, Q. Zheng, Q. Liu, and G. Pan, “Tinylight: Adaptive traffic signal control on devices with extremely limited resources,” in Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, IJCAI-22 , 7 2022, pp. 3999–4005. [Online]. Available:...

    2022 doi

  51. [59]

    Smart traffic control system using atmega328 micro controller and arduino software,

    B. N. Jyothi et al., “Smart traffic control system using atmega328 micro controller and arduino software,” in 2016 International Conference on Signal Processing, Communication, Power and Embedded System (SCOPES). IEEE, 2016, pp. 1584–1587

    2016

  52. [60]

    � -light: Programmatic interpretable reinforcement learning for resource-limited traffic signal control,

    Y . Gu, K. Zhang, Q. Liu, W. Gao, L. Li, and J. Zhou, “ � -light: Programmatic interpretable reinforcement learning for resource-limited traffic signal control,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 19, pp. 21 107–21 115, Mar. 2024. [Onli...

    2024

  53. [61]

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,

    S. Han, H. Mao, and W. J. Dally, “Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,” arXiv preprint arXiv:1510.00149 , 2015

    2015 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.