Pith. sign in

REVIEW 5 major objections 6 minor 53 references

Towards Efficient Few-shot Graph Neural Architecture Search via Partitioning Gradient Contribution

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper argues that the weight coupling problem in few-shot neural architecture search can be traced to conflicting gradient directions between adjacent-layer modules, and that a single Vector-Jacobian Product decomposition during…

desk verdict GC's VJP decomposition is a genuinely cheaper gradient-based partition method and the unified MPNN/GT search space is a useful step, but the min-cut formulation is underspecified for signed cosine weights and the headline partition-quality evidence is thinner than the abstract suggests. read the letter →

arxiv 2506.01231 v2 pith:GDTKVWWT submitted 2025-06-02 cs.LG cs.AI

Wenhao Song , Xuan Wu , Bo Yang , You Zhou , Yubin Xiao , Yanchun Liang , Hongwei Ge , Heow Pueh Lee
show 1 more author
Chunguo Wu
This is my paper
classification cs.LGcs.AI
keywords neuralarchitecturesearchfew-shotNASsupernetpartitioningweightcouplinggradientcontributiongraphmessagepassingnetworkstransformers
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 targets the weight coupling problem in few-shot neural architecture search, where a supernet's shared weights are used to judge subnets that will later be trained alone. It argues that the real cause is conflicting gradient directions: modules in one layer pull the shared weights of the preceding layer in opposite ways. To exploit this, it proposes Gradient Contribution (GC), which reads gradient directions off the supernet's normal backward pass with a decomposed Vector-Jacobian Product, and then uses a minimum cut over cosine similarities to send conflicting modules to different sub-supernets. Wrapped in the UGAS framework that searches message-passing and graph-transformer modules together, the paper reports that GC gives the best ranking correlation between inherited-weight and from-scratch evaluations, uses about 21 percent of the time of the gradient-matching baseline, and that UGAS+GC beats twelve hand-designed GNNs and other NAS methods on the benchmarks tested.

What carries the argument

The load-bearing object is the gradient contribution $C_{l,j} = (\partial y_{l,j} / \partial y_{l-1})^{\top} \nabla_{y_{l,j}} \mathcal{L}$, obtained by decomposing the Vector-Jacobian Product of a layer's backpropagation. It measures how each module in layer $l$ alone would push the weights of layer $l-1$. GC computes cosine similarities $S_{l,i,j}$ between these vectors, treats the modules as graph nodes with $S$ as edge weights, and applies the Stoer-Wagner minimum cut algorithm to split each layer into two sets; the top-$k$ lowest-cut layers are partitioned, generating $2^k$ sub-supernets. The argument that this works rests on a proof that a minimum-cut partition cannot leave both sides with below-average internal gradient similarity.

What would settle it

Train a supernet, record the GC partition from an early backward pass, then measure for many sampled subnets the accuracy gap between inherited-weight evaluation and from-scratch retraining; if the early-pass partition is no better than a random partition at reducing ranking error, or if partitions computed at later checkpoints give noticeably higher ranking correlation, the central assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that the weight-coupling error of a supernet can be predicted from one local quantity: the cosine similarity between the gradient contributions that adjacent-layer modules send back to the shared preceding layer. GC computes each module's gradient contribution $C_{l,j} = (\partial y_{l,j} / \partial y_{l-1})^{\top} \nabla_{y_{l,j}} \mathcal{L}$, forms a similarity graph per layer, and solves a minimum cut to split modules whose gradients conflict. Selecting the $k$ layers with the lowest cut weights yields $2^k$ sub-supernets, so each subnet inherits weights from a sub-supernet with less internal gradient conflict. The paper reports that this makes inherited weights track from-scratch performance more closely than existing few-shot partitioners, with Kendall and Spearman correlations of 0.54 and 0.73 on the Cluster dataset, and that architectures searched under UGAS outperform hand-designed GNNs and other NAS methods.

Load-bearing premise

The partitioning decision is made from gradient directions measured once in early supernet training, and the method assumes those early conflicts still predict which modules will fight each other when subnets are trained later.

Editorial extensions

If this is right

  • Few-shot NAS can rank subnets from inherited weights with less distortion: the reported 0.54 Kendall and 0.73 Spearman correlations on Cluster mean high-performing subnets are less likely to be discarded and poor ones less likely to be overestimated.
  • Partitioning cost drops to roughly one fifth of that of the gradient-matching baseline because no temporary supernets are pruned and retrained; the partition is read off the forward and backward passes already being run.
  • The search space can include structurally different modules, message-passing networks and graph transformers, so the search, not the search-space designer, decides which inductive biases each layer should use.
  • The partition logic is generic: any one-shot NAS search space with parallel modules in a layer could be split by gradient-direction conflicts, not just GNN backbones.
  • The ablation studies indicate $k=2$, giving four sub-supernets, is the reported sweet spot: deeper partitioning improves weight fidelity but shrinks module counts and raises search cost.

Reading between the lines

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

  • The appendix shows cut weights are stable across training stages, but that does not by itself prove early-stage gradient conflicts predict final subnet ranking divergence; a direct test would partition at a late checkpoint and compare the searched results.
  • Because GC only needs cosine similarities among modules within one layer, applying it to non-graph search spaces such as convolutional or transformer blocks is a natural extension the paper does not test.
  • The minimum-cut formulation treats each layer's partition independently; interactions between two partitioned layers could in principle cancel or amplify conflicts, so a joint multi-layer partition is a testable alternative.
  • Portability to other training regimes is untested: the ranking quality might depend on the particular optimizer, schedule, and early-training checkpoint used to collect the gradient contributions.
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

5 major / 6 minor

Summary. The paper proposes Gradient Contribution (GC), a supernet partitioning method for few-shot neural architecture search. GC decomposes the Vector-Jacobian Product during backpropagation to compute cosine similarities between gradient contributions of modules in adjacent layers, then formulates the partitioning of each layer as a minimum-cut problem over these similarities, selecting the top-k layers with the smallest cut weights to split the supernet into 2^k sub-supernets. The paper also introduces UGAS, a unified graph neural architecture search framework that searches over both MPNNs and graph transformers, with GA and DARTS implementations. Experiments on ten graph datasets claim that GC achieves state-of-the-art partitioning quality and time efficiency (about 21% of GM's time), and that UGAS+GC finds architectures outperforming manually designed GNNs and existing NAS methods.

Significance. If the claims hold, this is a valuable contribution to few-shot NAS: it offers a principled, gradient-based way to identify conflicting modules and partition a supernet with little extra training cost, and it broadens GNAS search spaces to include both MPNNs and graph transformers. The time-efficiency comparison against GM is clear and compelling, and the release of source code is a strength. However, the headline partitioning-quality claim rests on a single reduced dataset with overlapping error bars, and the minimum-cut formulation has a technical gap concerning signed edge weights. The unified MPNN/GT search space is a useful conceptual step, and the ablation in Table 6 supports the benefit of joint search.

major comments (5)
  1. [Section 4.4, Eq. (18)-(19), Algorithm 2] The cosine similarity S_{l,i,j} in Eq. (18) is signed, and Eq. (19) minimizes the sum of these signed weights over a cut. The Stoer-Wagner algorithm is only correct for nonnegative edge weights; with negative weights, the problem is not the standard global minimum-cut problem, and the algorithm's optimality guarantee does not apply. Figure 4 explicitly shows negative edge weights (e.g., -0.7, -0.5). The paper does not specify a transformation (shifting, clamping, or a balanced-cut formulation) that would make the optimization well-defined. Consequently, the partition scores gamma_l and the top-k layer selection are not necessarily actual minimizers of Eq. (19), and the SOTA partitioning-quality claim in Table 4 is not supported by the stated optimization. The authors should either prove their implementation solves Eq. (19) (e.g., by specifying the exact transformation in the released code) or reformulate the objective so that the solver is applicable.
  2. [Appendix B, Proposition 1] The proof of Proposition 1 assumes that the minimum-cut algorithm minimizes the cross-group similarity S_cross. Given the signed-weight issue in Section 4.4, this premise is not established. Moreover, the derivation of E[S_cross] in Eq. (29) is not fully justified (e.g., the probability expression and the treatment of self-similarities), and the conclusion that at least one group has above-average internal similarity is nearly definitional. The paper should either correct this proof to handle signed weights rigorously or present it as a heuristic motivation rather than a formal guarantee.
  3. [Section 5.3.2 and Appendix D, Table 4] The headline partitioning-quality result is based on a single dataset (Cluster, reduced to 30% of its original size), and the reported differences are within overlapping error bars (GC Kendall 0.54±0.17 vs. GM 0.42±0.22; Spearman 0.73±0.19 vs. 0.61±0.16). Given that the paper claims state-of-the-art partitioning quality, additional datasets or a statistical significance test are needed to support this claim. Without such evidence, the conclusion that GC reliably outperforms prior partitioning methods remains tentative.
  4. [Tables 2 and 3 (symbol †)] Many baseline numbers in Tables 2 and 3 are taken from the original papers (marked with †) without matched training protocols, while the UGAS results are obtained in this work. The claimed improvements over manually designed GNNs are modest (e.g., average accuracy improvements of 0.27% on BGNN), so cross-paper differences in training budgets, hyperparameters, and evaluation procedures could account for the gaps. The authors should either rerun the baselines under the same protocol for the key comparisons, or explicitly acknowledge and bound the impact of unmatched protocols on the claimed superiority.
  5. [Appendix F] The partition criterion is stated to use "early-stage gradients," but the exact checkpoint (epoch or iteration) at which the gradients are recorded is never specified. This is a free parameter that can affect the resulting partition and the reproducibility of the method. The paper should report the specific early-stage checkpoint used in all experiments and, ideally, show sensitivity of the final ranking correlations to this choice.
minor comments (6)
  1. [Section 3.2, Eq. (2)] The minimization in Eq. (2) is written as "min ∑ ..." without specifying the variable; it should read "min_{Γ⊆X} ∑ ..." for clarity.
  2. [Section 5.3.3 and Figure 5] The sentence "the supernet training time of GC is approximately four times that of GM" is confusing because Figure 5 shows that GC's total time is much smaller; clarify that the comparison refers to the supernet training component only, while the total partitioning time is what is summarized by the 21% figure.
  3. [Appendix F, Table 7] Table 7 reports ranking consistency values without standard deviations or the number of repetitions; adding this information would make the stability of the early-stage gradient claim more interpretable.
  4. [Figure 4] The similarity matrix in Figure 4 displays negative entries, which directly relates to the major concern about signing in the min-cut formulation; a brief note on how negative weights are handled in practice would help the reader.
  5. [Algorithm 2, line 3] The line "Compute the partition score γ_l" is vague; it should reference Eq. (19) and the specific solver used, especially given the signed-weight issue.
  6. [Abstract and Introduction] The paper uses "few-shot NAS" to describe methods that partition a supernet into sub-supernets; this terminology is used inconsistently in the literature, and a one-sentence clarification of the intended meaning would improve accessibility.

Circularity Check

1 steps flagged · score 4.0 of 10

The Appendix B 'proof of effectiveness' is definitional: it restates the min-cut objective rather than deriving weight-coupling mitigation, while the central empirical ranking-correlation claims remain independently measured.

  1. self definitional [Appendix B, Proposition 1 (Proof of the Effectiveness of GC)]
    "Proposition 1. There exists a subset Γ⊆M obtained by a gradient contribution-based minimum-cut algorithm such that the average gradient contribution similarity within Γ is greater than or equal to that of the original set:"

    This proposition is a direct restatement of the min-cut objective in Eq. (19): the algorithm is defined to minimize cross-group cosine similarity, and the conclusion that one side then has above-average internal similarity is a mathematical property of that objective, not a derivation that the partition mitigates weight coupling. The causal leap from gradient-contribution similarity to improved inherited-weight fidelity is assumed in the problem definition and by the citation to prior work, so the 'proof of effectiveness' reduces to the method's own construction. The empirical ranking-correlation experiments are independent and are not forced by this step, but the theoretical justification itself is self-definitional.

full rationale

The central empirical pipeline is self-contained and not circular: GC decomposes the VJP of the trained supernet's gradients (Eqs. 16-18), forms cosine-similarity edge weights, solves a min-cut, retrains sub-supernets, and then measures Kendall/Spearman correlations against independent from-scratch training. These measurements are external to the construction and could in principle disfavor the method; no parameter is fitted to the ranking-correlation target. The only genuinely definitional move is the Appendix B proof, which proves a property of the min-cut objective rather than the intended claim that grouping by gradient similarity reduces weight coupling. The signed-edge Stoer-Wagner concern is a correctness/soundness issue (Stoer-Wagner assumes nonnegative weights), not a circularity. The choice of k=2 is tuned on Cluster and then used in the Cluster ranking-correlation table, which is mild selection rather than circularity. Self-citations in the reference list are background and not load-bearing. Overall, the empirical claims have independent content, but the theoretical 'effectiveness proof' is partially circular by construction, giving a moderate score.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The paper introduces no physical or external entities; its new objects are algorithmic (GC and UGAS). The central claim rests on the domain assumptions listed above, especially the early-stage gradient proxy and the min-cut criterion, plus standard automatic differentiation and min-cut machinery.

free parameters (3)
  • k (number of partition points) = 2
    Set to 2 (four sub-supernets) across all datasets; selected based on the Cluster ablation in Table 5, which shows k=2 optimal.
  • MPGT layers L per dataset = 4 to 10, 15 for Cluster
    Layer count is tuned per dataset; Table 8 uses a Cluster layer-number sweep to pick 15. Affects capacity and search cost.
  • Early-stage gradient checkpoint = unspecified
    Appendix F says early-stage gradients are used for partitioning, but the exact epoch or step is not stated; this choice affects the similarity matrix and the resulting partition.
assumptions (7)
  • domain assumption Weight coupling is primarily caused by conflicting gradient directions from succeeding-layer modules to preceding-layer modules.
    Core premise behind GC, stated in Section 1 and Section 4.3; not derived.
  • domain assumption Cosine similarity of gradient contributions is a valid proxy for harmful gradient conflict.
    Used in Eq. (18) to build edge weights for the minimum cut; no theoretical justification beyond intuition.
  • domain assumption Early-stage gradient conflicts remain representative through training.
    Appendix F reports high ranking consistency between stages (Table 7), but this is empirical on limited settings, not a proof.
  • domain assumption Minimum-cut partition based on these similarities improves subnet weight reliability.
    Appendix B proves a group-average similarity inequality, not that inherited sub-supernet weights match from-scratch training.
  • domain assumption MPNN and GT outputs can be combined per layer by averaging and MLP fusion.
    Fusion in Eqs. (5)-(7) is a design choice with no theoretical guarantee; validated by ablations only.
  • standard math The chain rule and VJP decomposition correctly attribute gradients to individual modules.
    Standard automatic differentiation; Eqs. (13)-(16).
  • standard math The Stoer-Wagner algorithm returns a minimum cut.
    Classic result cited as [36].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Efficient Few-shot Graph Neural Architecture Search via Partitioning Gradient Contribution." pith.science (2026). https://pith.science/paper/GDTKVWWT

@misc{pith2026250601231,
  author       = {Pith},
  title        = {Pith review of: Towards Efficient Few-shot Graph Neural Architecture Search via Partitioning Gradient Contribution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GDTKVWWT}},
  note         = {Machine review of arXiv:2506.01231}
}
read the original abstract

To address the weight coupling problem, certain studies introduced few-shot Neural Architecture Search (NAS) methods, which partition the supernet into multiple sub-supernets. However, these methods often suffer from computational inefficiency and tend to provide suboptimal partitioning schemes. To address this problem more effectively, we analyze the weight coupling problem from a novel perspective, which primarily stems from distinct modules in succeeding layers imposing conflicting gradient directions on the preceding layer modules. Based on this perspective, we propose the Gradient Contribution (GC) method that efficiently computes the cosine similarity of gradient directions among modules by decomposing the Vector-Jacobian Product during supernet backpropagation. Subsequently, the modules with conflicting gradient directions are allocated to distinct sub-supernets while similar ones are grouped together. To assess the advantages of GC and address the limitations of existing Graph Neural Architecture Search methods, which are limited to searching a single type of Graph Neural Networks (Message Passing Neural Networks (MPNNs) or Graph Transformers (GTs)), we propose the Unified Graph Neural Architecture Search (UGAS) framework, which explores optimal combinations of MPNNs and GTs. The experimental results demonstrate that GC achieves state-of-the-art (SOTA) performance in supernet partitioning quality and time efficiency. In addition, the architectures searched by UGAS+GC outperform both the manually designed GNNs and those obtained by existing NAS methods. Finally, ablation studies further demonstrate the effectiveness of all proposed methods.

Figures

Figures reproduced from arXiv: 2506.01231 by the authors.

Figure 1
Figure 1. (A) illustrates the process of sampling a subnet from [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the supernet architecture. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Illustration of the GC-based supernet partitioning [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Comparison of supernet partitioning time between [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Illustration of the differences between GC and GM. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the Searched Architecture on the [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 46 canonical work pages

  1. [1]

    Uri Alon and Eran Yahav. 2021. On the Bottleneck of Graph Neural Networks and its Practical Implications. InProceedings of the International Conference on Learning Representations

    work page 2021

  2. [2]

    Pearlmutter, Alexey Andreyevich Radul, and Jeffrey Mark Siskind

    Atılım Günes Baydin, Barak A. Pearlmutter, Alexey Andreyevich Radul, and Jeffrey Mark Siskind. 2017. Automatic differentiation in machine learning: a survey.Journal of Machine Learning Research18, 1 (2017), 5595–5637

    work page 2017

  3. [3]

    Dominique Beaini, Saro Passaro, Vincent Létourneau, Will Hamilton, Gabriele Corso, and Pietro Lió. 2021. Directional Graph Networks. InProceedings of the International Conference on Machine Learning. 748–758

    work page 2021

  4. [4]

    Xavier Bresson and Thomas Laurent. 2017. Residual Gated Graph ConvNets. arXiv: 1711.07553

    arXiv 2017

  5. [5]

    Dexiong Chen, Leslie O’Bray, and Karsten Borgwardt. 2022. Structure-Aware Transformer for Graph Representation Learning. InProceedings of the Interna- tional Conference on Machine Learning. 3469–3489

    work page 2022

  6. [6]

    Lin Chen, Fengli Xu, Nian Li, Zhenyu Han, Meng Wang, Yong Li, and Pan Hui

    work page

  7. [7]

    Yuncheol Choi, Sun Woo Park, Minho Lee, and Youngho Woo. 2024. Topology- Informed Graph Transformer. arXiv:2402.02005

    arXiv 2024

  8. [8]

    Colwell, and Adrian Weller

    Krzysztof Marcin Choromanski, Valerii Likhosherstov, David Dohan, Xingyou Song, Andreea Gane, Tamás Sarlós, Peter Hawkins, Jared Quincy Davis, Afroz Mohiuddin, Lukasz Kaiser, David Benjamin Belanger, Lucy J. Colwell, and Adrian Weller. 2021. Rethinking Attention with Performers. InProceedings of the Inter- national Conference on Learning Representations

    work page 2021

Show all 53 references
  1. [9]

    Gabriele Corso, Luca Cavalleri, Dominique Beaini, Pietro Liò, and Petar Veličković

  2. [10]

    Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson

    Vijay Prakash Dwivedi, Chaitanya K. Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. 2023. Benchmarking Graph Neural Networks. Journal of Machine Learning Research24 (2023), 1–48

    2023

  3. [11]

    Vijay Prakash Dwivedi, Ladislav Rampášek, Michael Galkin, Ali Parviz, Guy Wolf, Anh Tuan Luu, and Dominique Beaini. 2022. Long Range Graph Benchmark. InProceedings of the Advances in Neural Information Processing Systems. 22326– 22340

    2022

  4. [12]

    Schoenholz, Patrick F

    Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. 2017. Neural Message Passing for Quantum Chemistry. InProceedings of the International Conference on Machine Learning. 1263–1272

    2017

  5. [13]

    Zichao Guo, Xiangyu Zhang, Haoyuan Mu, Wen Heng, Zechun Liu, Yichen Wei, and Jian Sun. 2020. Single Path One-Shot Neural Architecture Search with Uniform Sampling. InProceedings of the European Conference on Computer Vision. 544–560

    2020

  6. [14]

    Xin He, Jiangchao Yao, Yuxin Wang, Zhenheng Tang, Ka Chun Cheung, Simon See, Bo Han, and Xiaowen Chu. 2023. NAS-LID: Efficient Neural Architecture Search with Local Intrinsic Dimension. InProceedings of the AAAI Conference on Artificial Intelligence. 7839–7847

    2023

  7. [15]

    Hirsch and Stephen Smale

    Morris W. Hirsch and Stephen Smale. 1974.Differential Equations, Dynamical Systems, and Linear Algebra. Vol. 60. Academic Press, New York. 1–359 pages

    1974

  8. [16]

    Shoukang Hu, Ruochen Wang, Lanqing Hong, Zhenguo Li, Cho-Jui Hsieh, and Jiashi Feng. 2022. Generalizing Few-Shot NAS with Gradient Matching. In Proceedings of the International Conference on Learning Representations

    2022

  9. [17]

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open Graph Benchmark: Datasets for Machine Learning on Graphs. InProceedings of the Advances in Neural Information Processing Systems. 22118–22133

    2020

  10. [18]

    Pande, and Jure Leskovec

    Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay S. Pande, and Jure Leskovec. 2020. Strategies for Pre-training Graph Neural Net- works. InProceedings of the International Conference on Learning Representations

    2020

  11. [19]

    Zaki, and Dharmashankar Subramanian

    Md Shamim Hussain, Mohammed J. Zaki, and Dharmashankar Subramanian. 2022. Global Self-Attention as a Replacement for Graph Convolution. InProceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 655–665

    2022

  12. [20]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with Graph Convolutional Networks. InProceedings of the International Conference on Learning Representations

    2017

  13. [21]

    Devin Kreuzer, Dominique Beaini, Will Hamilton, Vincent Létourneau, and Pru- dencio Tossou. 2021. Rethinking Graph Transformers with Spectral Attention. InProceedings of the Advances in Neural Information Processing Systems. 21618– 21629

    2021

  14. [22]

    Guihong Li, Duc Hoang, Kartikeya Bhardwaj, Ming Lin, Zhangyang Wang, and Radu Marculescu. 2024. Zero-Shot Neural Architecture Search: Challenges, Solutions, and Opportunities.IEEE Transactions on Pattern Analysis and Machine Intelligence46, 12 (2024), 7618–7635

    2024

  15. [23]

    Thabet, and Bernard Ghanem

    Guohao Li, Chenxin Xiong, Ali K. Thabet, and Bernard Ghanem. 2020. Deep- erGCN: All You Need to Train Deeper GCNs. arXiv:2006.07739

    2020 arXiv

  16. [24]

    Hanxiao Liu, Karen Simonyan, and Yiming Yang. 2019. DARTS: Differentiable Architecture Search. InProceedings of the International Conference on Learning Representations

    2019

  17. [25]

    Doka- nia, Mark Coates, Philip Torr, and Ser-Nam Lim

    Liheng Ma, Chen Lin, Derek Lim, Adriana Romero-Soriano, Puneet K. Doka- nia, Mark Coates, Philip Torr, and Ser-Nam Lim. 2023. Graph Inductive Biases in Transformers without Message Passing. InProceedings of the International Conference on Machine Learning. 23321–23337

    2023

  18. [26]

    Youngmin Oh, Hyunju Lee, and Bumsub Ham. 2025. Efficient Few-Shot Neu- ral Architecture Search by Counting the Number of Nonlinear Functions. In Proceedings of the AAAI Conference on Artificial Intelligence

    2025

  19. [27]

    Kenta Oono and Taiji Suzuki. 2019. On Asymptotic Behaviors of Graph CNNs from Dynamical Systems Perspective. arXiv:1905.10947

    2019 arXiv

  20. [28]

    Guan, Barret Zoph, Quoc V

    Hieu Pham, Melody Y. Guan, Barret Zoph, Quoc V. Le, and Jeff Dean. 2018. Efficient Neural Architecture Search via Parameter Sharing. InProceedings of the International Conference on Machine Learning. 4092–4101

    2018

  21. [29]

    Aloïs Pourchot, Alexis Ducarouge, and Olivier Sigaud. 2020. To Share or Not To Share: A Comprehensive Appraisal of Weight-Sharing. arXiv:2002.04289

    2020 arXiv

  22. [30]

    Yijian Qin, Xin Wang, Ziwei Zhang, Hong Chen, and Wenwu Zhu. 2023. Multi- task graph neural architecture search with task-aware collaboration and cur- riculum. InProceedings of the Advances in Neural Information Processing Systems. 24879–24891

    2023

  23. [31]

    Yijian Qin, Xin Wang, Ziwei Zhang, Pengtao Xie, and Wenwu Zhu. 2022. Graph Neural Architecture Search Under Distribution Shifts. InProceedings of the Inter- national Conference on Machine Learning. 18083–18095

    2022

  24. [32]

    Zhengzhong Qiu, Wei Bi, Dong Xu, Hua Guo, Hongwei Ge, Yanchun Liang, Heow Pueh Lee, and Chunguo Wu. 2023. Efficient Self-learning Evolutionary Neural Architecture Search.Applied Soft Computing146 (2023), 110671

    2023

  25. [33]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer.Machine Learning Research21, 140 (2020), 1–67. KDD ’25, August ...

    2020

  26. [34]

    Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. 2022. Recipe for a General, Powerful, Scalable Graph Transformer. InProceedings of the Advances in Neural Information Processing Systems. 14501–14515

    2022

  27. [35]

    Esteban Real, Alok Aggarwal, Yanping Huang, and Quoc V. Le. 2019. Regularized Evolution for Image Classifier Architecture Search. InProceedings of the AAAI Conference on Artificial Intelligence. 4780–4789

    2019

  28. [36]

    Mechthild Stoer and Frank Wagner. 1997. A simple min-cut algorithm.J. ACM 44, 4 (1997), 585–591

    1997

  29. [37]

    Bronstein

    Jake Topping, Francesco Di Giovanni, Benjamin Paul Chamberlain, Xiaowen Dong, and Michael M. Bronstein. 2022. Understanding Over-Squashing and Bot- tlenecks on Graphs via Curvature. InProceedings of the International Conference on Learning Representations

    2022

  30. [38]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention Is All You Need. InProceedings of the Advances in Neural Information Processing Sys- tems. 5998–6008

    2017

  31. [39]

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. InProceedings of the International Conference on Learning Representations

    2018

  32. [40]

    Lanning Wei, Huan Zhao, Quanming Yao, and Zhiqiang He. 2021. Pooling Archi- tecture Search for Graph Classification. InProceedings of the ACM International Conference on Information and Knowledge Management. 2091––2100

    2021

  33. [41]

    Xuan Wu, Jizong Han, Di Wang, Pengyue Gao, Quanlong Cui, Liang Chen, Yanchun Liang, Han Huang, Heow Pueh Lee, Chunyan Miao, You Zhou, and Chunguo Wu. 2023. Incorporating Surprisingly Popular Algorithm and Euclidean distance-based adaptive topology into PSO.Swarm and Evolutiona...

    2023

  34. [42]

    Xuan Wu, Di Wang, Huanhuan Chen, Lele Yan, Yubin Xiao, Chunyan Miao, Hongwei Ge, Dong Xu, Yanchun Liang, Kangping Wang, Chunguo Wu, and You Zhou. 2024. Neural Architecture Search for Text Classification With Limited Computing Resources Using Efficient Cartesian Genetic Program...

    2024

  35. [43]

    Wright, Azalia Mirhoseini, Joseph E

    Zhanghao Wu, Paras Jain, Matthew A. Wright, Azalia Mirhoseini, Joseph E. Gonzalez, and Ion Stoica. 2021. Representing Long-Range Context for Graph Neural Networks with Global Attention. InProceedings of the Advances in Neural Information Processing Systems. 13266–13279

    2021

  36. [44]

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural Networks?. InProceedings of the International Conference on Learning Representations

    2019

  37. [45]

    Yang Yao, Xin Wang, Yi Qin, Ziwei Zhang, Wenwu Zhu, and Hong Mei. 2024. Data- Augmented Curriculum Graph Neural Architecture Search under Distribution Shifts. InProceedings of the AAAI Conference on Artificial Intelligence. 16433– 16441

    2024

  38. [46]

    Zizhao Zhang, Xin Wang, Chaoyu Guan, Ziwei Zhang, Haoyang Li, and Wenwu Zhu. 2023. AutoGT: Automated Graph Transformer Architecture Search. In Proceedings of the International Conference on Learning Representations

    2023

  39. [47]

    Yiyang Zhao, Linnan Wang, Yuandong Tian, Rodrigo Fonseca, and Tian Guo

  40. [48]

    Xun Zhou, Xingyu Wu, Liang Feng, Zhichao Lu, and Kay Chen Tan. 2024. Design Principle Transfer in Neural Architecture Search via Large Language Models. arxiv: 2408.11330

    2024 arXiv

  41. [49]

    Barret Zoph and Quoc V. Le. 2017. Neural Architecture Search with Reinforcement Learning. InProceedings of the International Conference on Learning Representa- tions

    2017

  42. [50]

    Barret Zoph, Vijay Vasudevan, Jonathon Shlens, and Quoc V. Le. 2018. Learning Transferable Architectures for Scalable Image Recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 8697–8710. A Detailed Parameter Settings In this section, w...

    2018

  43. [2020]

    InProceedings of the Advances in Neural Information Processing Systems

    Principal Neighbourhood Aggregation for Graph Nets. InProceedings of the Advances in Neural Information Processing Systems. 13260–13271

  44. [2021]

    InProceedings of the International Conference on Machine Learning

    Few-Shot Neural Architecture Search. InProceedings of the International Conference on Machine Learning. 12707–12718

  45. [2024]

    InProceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Large Language Model-driven Meta-structure Discovery in Heteroge- neous Information Network. InProceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 307–318

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.