pith. sign in

arxiv: 2606.22429 · v1 · pith:4J27JLYXnew · submitted 2026-06-21 · 💻 cs.LG

Enhancing LLMs for Graph Tasks via Graph-aware LoRA Generation

Pith reviewed 2026-06-26 10:57 UTC · model grok-4.3

classification 💻 cs.LG
keywords Graph-aware adaptationLow-rank adaptationLarge language modelsGraph tasksZero-shot learningWeight generationWhole-graph information
0
0 comments X

The pith

GaRA generates low-rank weight updates conditioned on graph structures to inject whole-graph information into LLMs.

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

The paper proposes a weight-level information injection paradigm to adapt language models to graph tasks by generating task-specific weight updates that interact directly with hidden representations. Existing methods lose whole-graph information during adaptation, but GaRA instantiates the paradigm via low-rank adaptation by building updates from the original graph and constraining their norms. This approach aims to encode complete graph structure into the model without the optimization bias seen in prior techniques. Empirical results show consistent outperformance on zero-shot graph learning tasks.

Core claim

GaRA constructs low-rank weight updates conditioned on the original graph structures and constrains the norm of the generated updates, thus injecting whole-graph information and avoiding the optimization bias in the weight generation.

What carries the argument

GaRA, the Graph-aware LoRA generation model that builds task-specific low-rank weight updates directly from input graph structures.

If this is right

  • GaRA enables language models to handle graph tasks with better transferability across datasets than graph neural networks.
  • The norm constraint prevents optimization bias during generation of the weight updates.
  • Whole-graph information reaches the model's hidden representations through direct interaction with the generated updates.
  • Performance improves consistently on zero-shot graph learning tasks compared to existing adaptation baselines.

Where Pith is reading between the lines

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

  • The same conditioning approach could extend to other low-rank adaptation variants beyond LoRA.
  • If the paradigm holds, it suggests a general route for injecting structured data modalities into frozen language models without full retraining.

Load-bearing premise

Generating and applying low-rank weight updates conditioned on graph structure will successfully encode whole-graph information into the LLM's hidden representations without the information loss seen in prior adaptation methods.

What would settle it

A controlled test where GaRA-adapted LLMs still fail to distinguish isomorphic graphs or show no gain over standard LoRA on whole-graph classification metrics.

Figures

Figures reproduced from arXiv: 2606.22429 by Junshu Sun, Qingming Huang, Shuhui Wang, Wanxing Chang.

Figure 1
Figure 1. Figure 1: Comparison between Graph Injection Paradigms. (a) Output-level injection paradigm employs GNNs as the model predictor, failing to tackle graphs with diverse labels. (b) Input￾level injection paradigm transforms input graphs into subgraph sequences, overlooking higher-level information in the original graphs and leading to potential information loss. (c) Our weight￾level injection paradigm enables LMs to le… view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline of the Graph-aware LoRA Generation model, GaRA. (a) To inject whole-graph information into LLMs, GaRA first applies GNN to the whole input graph, extracting a global representation via max pooling. This representation is then fed into an MLP-based weight generator to obtain low-rank weight updates G. (b) To tackle the norm-amplification optimization bias, the norm of G is further rescaled based on… view at source ↗
Figure 3
Figure 3. Figure 3: Effectiveness in Tackling Norm Amplification. “nr" denotes our norm rescaling strategy. (a) and (b) The lower figures are the zoom-in illustrations of the upper figures. (c) “Similarity" denotes the cosine similarity between the updated weight W and the original weight Wˆ . Results are averaged over the 50 randomly selected test samples, where the selection is refreshed every 50 training steps. better perf… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison Among Injection Paradigms. “Sub￾Graph" solely employs subgraph features as input. “Whole" further concatenates global features to the input sequences. 1 11 21 31 32 Depth of the Injected Layer 40 42 44 46 48 50 Average Model Performance Average Left Matrix Right Matrix [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison Among Generation Inputs. “Random", “Non-Graph", and "SubGraph" denote generating low-rank ma￾trices from random vectors, embedded task description, and GNN-processed subgraph features, respectively. CoRA WikiCS Reddit Instagram 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Model Performance 0.0004 0.0009 0.0023 0.0017 0.2355 0.4276 0.3792 0.3081 Whole (GaRA) SubGraph [PITH_FULL_IMAGE:figures/ful… view at source ↗
read the original abstract

Graph neural networks (GNNs) tightly couple their input-output parameters to dataset-specific feature spaces and target sets, exhibiting limited transferability across different datasets. In contrast, language models (LMs) generalize flexibly via a unified input-output interface, motivating recent attempts to adapt LMs to graph tasks. However, existing methods struggle to encode whole-graph information, leading to potential information loss and suboptimal graph understanding. In this work, we propose a novel weight-level information injection paradigm for adapting LMs to graph tasks. This paradigm injects whole-graph information by generating task-specific weight updates that interact directly with hidden representations. Instantiating this paradigm following low-rank adaptation (LoRA), we introduce GaRA, a Graph-aware LoRA generation model. GaRA constructs low-rank weight updates conditioned on the original graph structures and constrains the norm of the generated updates, thus injecting whole-graph information and avoiding the optimization bias in the weight generation. Empirical studies demonstrate that GaRA consistently outperforms baselines on zero-shot graph learning tasks.

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 / 1 minor

Summary. The manuscript proposes a weight-level information injection paradigm for adapting language models to graph tasks. It instantiates this as GaRA, which generates low-rank (LoRA) weight updates conditioned on the input graph structure and applies a norm constraint on the generated updates. The central claim is that this construction injects whole-graph information into the LLM hidden representations while avoiding the information loss and optimization bias of prior adaptation methods, leading to consistent outperformance on zero-shot graph learning tasks.

Significance. If the empirical claims hold under rigorous evaluation, the work would be significant for the graph+LLM literature by offering a concrete mechanism for direct whole-graph conditioning at the parameter-update level rather than through input augmentation or prompt engineering. The norm-constrained generation step is a potentially reusable design choice that could be tested on other adaptation settings.

major comments (2)
  1. [Abstract, §3] Abstract and §3 (method description): the claim that the norm constraint 'avoids the optimization bias in the weight generation' is stated at a high level without a derivation, ablation, or analysis showing how the constraint interacts with the graph-conditioned generator to preserve whole-graph information. This is load-bearing for the central mechanistic claim.
  2. [Abstract] Abstract: the statement of 'consistent outperformance on zero-shot graph learning tasks' is presented without any quantitative results, dataset names, baseline descriptions, or statistical details. Because the soundness of the empirical support cannot be assessed, the central claim that GaRA solves the information-loss problem remains unevaluated.
minor comments (1)
  1. [§3] Notation for the graph-conditioned generator and the norm constraint should be introduced with explicit equations rather than prose descriptions to allow readers to verify the construction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major comment below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses
  1. Referee: [Abstract, §3] Abstract and §3 (method description): the claim that the norm constraint 'avoids the optimization bias in the weight generation' is stated at a high level without a derivation, ablation, or analysis showing how the constraint interacts with the graph-conditioned generator to preserve whole-graph information. This is load-bearing for the central mechanistic claim.

    Authors: We agree that the current presentation of the norm constraint's role is high-level. In the revised manuscript we will expand §3 to include a derivation of how the norm constraint regularizes the graph-conditioned generator, preventing collapse to solutions that discard structural information, and add an ablation study (with and without the constraint) to quantify its contribution to whole-graph information preservation. revision: yes

  2. Referee: [Abstract] Abstract: the statement of 'consistent outperformance on zero-shot graph learning tasks' is presented without any quantitative results, dataset names, baseline descriptions, or statistical details. Because the soundness of the empirical support cannot be assessed, the central claim that GaRA solves the information-loss problem remains unevaluated.

    Authors: We acknowledge that the abstract's high-level claim would benefit from concrete support for evaluation. We will revise the abstract to include key quantitative highlights (e.g., average gains, dataset names, and main baselines) while respecting length constraints, thereby allowing direct assessment of the empirical claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper introduces GaRA as a novel weight-level injection paradigm for adapting LMs to graphs via graph-conditioned LoRA generation with norm constraints. The provided abstract and description contain no equations, derivations, or load-bearing self-citations. The central claim is presented as a design choice supported by empirical outperformance on zero-shot tasks, without any reduction of predictions or uniqueness results to fitted inputs or prior author work by construction. The derivation chain is self-contained at the level of high-level method description.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be extracted beyond the high-level description of the new paradigm.

axioms (1)
  • domain assumption Whole-graph information can be injected via generated weight updates that interact with hidden representations
    Core premise of the proposed paradigm stated in the abstract
invented entities (1)
  • GaRA no independent evidence
    purpose: Graph-aware LoRA generation model that conditions updates on graph structures
    New model introduced to instantiate the weight-level injection paradigm

pith-pipeline@v0.9.1-grok · 5708 in / 1100 out tokens · 28686 ms · 2026-06-26T10:57:13.367714+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

299 extracted references · 68 canonical work pages · 5 internal anchors

  1. [1]

    Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

    Reimers, Nils and Gurevych, Iryna. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. 2019

  2. [2]

    Meta AI , urldate =

    Introducing. Meta AI , urldate =

  3. [3]

    Visual Perception by Large Language Model's Weights , booktitle =

    Ma, Feipeng and Xue, Hongwei and Zhou, Yizhou and Wang, Guangting and Rao, Fengyun and Yan, Shilin and Zhang, Yueyi and Wu, Siying and Shou, Mike Zheng and Sun, Xiaoyan , year = 2024, volume =. Visual Perception by Large Language Model's Weights , booktitle =

  4. [4]

    A Universal Approximation Theorem of Deep Neural Networks for Expressing Probability Distributions , booktitle =

    Lu, Yulong and Lu, Jianfeng , year =. A Universal Approximation Theorem of Deep Neural Networks for Expressing Probability Distributions , booktitle =

  5. [5]

    2023 , publisher=

    Matrix analysis and applied linear algebra , author=. 2023 , publisher=

  6. [6]

    Instance

    Fortin, Jean-Michel and Gamache, Olivier and Grondin, Vincent and Pomerleau, Fran. Instance. 2022 , pages =

  7. [7]

    International Journal of Computer Vision , volume =

    Zhang, Libo and Jiang, Lutao and Ji, Ruyi and Fan, Heng , year =. International Journal of Computer Vision , volume =

  8. [8]

    , year =

    Chen, Mark and Tworek, Jerry and Jun, Heewoo et al. , year =. Evaluating. 2107.03374 , primaryclass =

  9. [9]

    Cordts, Marius and Omran, Mohamed and Ramos, Sebastian and Rehfeld, Timo and Enzweiler, Markus and Benenson, Rodrigo and Franke, Uwe and Roth, Stefan and Schiele, Bernt , year =. The

  10. [10]

    2009 , pages =

    Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and. 2009 , pages =

  11. [11]

    Measuring

    Hendrycks, Dan and Burns, Collin and Basart, Steven and Zou, Andy and Mazeika, Mantas and Song, Dawn and Steinhardt, Jacob , year =. Measuring. International

  12. [12]

    Diffusion

    Klicpera, Johannes and. Diffusion. International. 2019 , volume =

  13. [13]

    Adaptive

    Chien, Eli and Peng, Jianhao and Li, Pan and Milenkovic, Olgica , year =. Adaptive. International

  14. [14]

    and Kaiser,

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N. and Kaiser,. Attention Is All You Need , booktitle =. 2017 , pages =

  15. [15]

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian , year =. Deep

  16. [16]

    , year =

    Krizhevsky, Alex and Sutskever, Ilya and Hinton, Geoffrey E. , year =. Communications of the ACM , volume =

  17. [17]

    Predict Then

    Gasteiger, Johannes and Bojchevski, Aleksandar and G. Predict Then. International. 2018 , address =

  18. [18]

    Simplifying

    Wu, Felix and Souza, Amauri and Zhang, Tianyi and Fifty, Christopher and Yu, Tao and Weinberger, Kilian , year =. Simplifying. International

  19. [19]

    Huang, Zhongyu and Wang, Yingheng and Li, Chaozhuo and He, Huiguang , year =. Going. International

  20. [20]

    and Tang, Jian , year =

    Zhao, Jianan and Zhu, Zhaocheng and Galkin, Mikhail and Mostafa, Hesham and Bronstein, Michael M. and Tang, Jian , year =. Fully-Inductive. The

  21. [21]

    Proceedings of the

    Hou, Zhenyu and He, Yufei and Cen, Yukuo and Liu, Xiao and Dong, Yuxiao and Kharlamov, Evgeny and Tang, Jie , year =. Proceedings of the. doi:10.1145/3543507.3583379 , urldate =

  22. [22]

    and Ventola, Pamela and Duncan, James S

    Li, Xiaoxiao and Zhou, Yuan and Dvornek, Nicha and Zhang, Muhan and Gao, Siyuan and Zhuang, Juntang and Scheinost, Dustin and Staib, Lawrence H. and Ventola, Pamela and Duncan, James S. , year =. Medical Image Analysis , volume =

  23. [23]

    Li, Jintang and Wu, Ruofan and Sun, Wangbin and Chen, Liang and Tian, Sheng and Zhu, Liang and Meng, Changhua and Zheng, Zibin and Wang, Weiqiang , year =. What's. Proceedings of the 29th. doi:10.1145/3580305.3599546 , urldate =

  24. [24]

    , year =

    Xia, Jun and Wu, Lirong and Chen, Jintao and Hu, Bozhen and Li, Stan Z. , year =. Proceedings of the. doi:10.1145/3485447.3512156 , urldate =

  25. [25]

    Graph Contrastive Learning with Augmentations , booktitle =

    You, Yuning and Chen, Tianlong and Sui, Yongduo and Chen, Ting and Wang, Zhangyang and Shen, Yang , year =. Graph Contrastive Learning with Augmentations , booktitle =

  26. [26]

    Yu, Xingtong and Gong, Zechuan and Zhou, Chang and Fang, Yuan and Zhang, Hui , year =

  27. [27]

    Zhao, Jitao and Jin, Di and Ge, Meng and Shan, Lianze and Wang, Xin and He, Dongxiao and Feng, Zhiyong , year =. The

  28. [28]

    2019 , pages =

    International. 2019 , pages =

  29. [29]

    2017 , journal =

    The Geometric Nature of Weights in Real Complex Networks , author =. 2017 , journal =

  30. [30]

    Alon, Uri and Yahav, Eran , year =. On the. International. arxiv , langid =:2006.05205 , address =

  31. [31]

    Subgraph Neural Networks , booktitle =

    Alsentzer, Emily and Finlayson, Samuel and Li, Michelle and Zitnik, Marinka , editor =. Subgraph Neural Networks , booktitle =. 2020 , volume =

  32. [32]

    Learnable

    Anonymous , year =. Learnable. Submitted to

  33. [33]

    2008 , journal =

    Synchronization in Complex Networks , author =. 2008 , journal =

  34. [34]

    Synchronization

    Arenas, Alex and. Synchronization. 2006 , journal =

  35. [35]

    Arora, Siddhant , year =. A. arXiv:2007.12374 [cs] , eprint =

  36. [36]

    Arvind, Vikraman and Fuhlbr. On. Fundamentals of. 2019 , pages =

  37. [37]

    Advances in Neural Information Processing Systems , author =

    Diffusion-. Advances in Neural Information Processing Systems , author =. 2016 , eprint =

  38. [38]

    Expressive

    Azizian, Wa. Expressive. 2021 , journal =. 2006.15646 , primaryclass =

  39. [39]

    Breaking the

    Balcilar, Muhammet and Heroux, Pierre and Gauzere, Benoit and Vasseur, Pascal and Adam, Sebastien and Honeine, Paul , year =. Breaking the. Proceedings of the 38th

  40. [40]

    Bridging the

    Balcilar, Muhammet and Renton, Guillaume and Heroux, Pierre and Gauzere, Benoit and Adam, Sebastien and Honeine, Paul , year =. Bridging the. arXiv:2003.11702 [cs, stat] , eprint =

  41. [41]

    Barcel. Graph. Advances in. 2021 , volume =

  42. [42]

    Bastos, Anson and Nadgeri, Abhishek and Singh, Kuldeep and Kanezashi, Hiroki and Suzumura, Toyotaro and Mulang', Isaiah Onando , year =. How. Transactions on Machine Learning Research , issn =

  43. [43]

    Networks beyond Pairwise Interactions:

    Battiston, Federico and Cencetti, Giulia and Iacopini, Iacopo and Latora, Vito and Lucas, Maxime and Patania, Alice and Young, Jean-Gabriel and Petri, Giovanni , year =. Networks beyond Pairwise Interactions:. Physics Reports , volume =

  44. [44]

    Directional

    Beaini, Dominique and Passaro, Saro and L. Directional. 2020 , journal =. 2010.02863 , primaryclass =

  45. [45]

    Laplacian

    Belkin, Mikhail and Niyogi, Partha , year =. Laplacian. Neural Computation , volume =

  46. [46]

    and Bahdanau, Dzmitry , year =

    Bergen, Leon and O'Donnell, Timothy J. and Bahdanau, Dzmitry , year =. Systematic. Advances in

  47. [47]

    Bessadok, Alaa and Mahjoub, Mohamed Ali and Rekik, Islem , year =. Graph. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

  48. [48]

    and Maron, Haggai , year =

    Bevilacqua, Beatrice and Frasca, Fabrizio and Lim, Derek and Srinivasan, Balasubramaniam and Cai, Chen and Balamurugan, Gopinath and Bronstein, Michael M. and Maron, Haggai , year =. Equivariant. International

  49. [49]

    Bi, Wendong and Du, Lun and Fu, Qiang and Wang, Yanlin and Han, Shi and Zhang, Dongmei , year =. Make

  50. [50]

    Specformer:

    Bo, Deyu and Shi, Chuan and Wang, Lele and Liao, Renjie , year =. Specformer:. The

  51. [51]

    and Latora, V

    Boccaletti, S. and Latora, V. and Moreno, Y. and Chavez, M. and Hwang, D. -U. , year =. Complex Networks:. Physics Reports , volume =

  52. [52]

    Weisfeiler and

    Bodnar, Cristian and Frasca, Fabrizio and Otter, Nina and Wang, Yu Guang and Li. Weisfeiler and. Advances in. 2021 , urldate =

  53. [53]

    Bodnar, Cristian and Frasca, Fabrizio and Wang, Yuguang and Otter, Nina and Montufar, Guido F. and Li. Weisfeiler and. Proceedings of the 38th. 2021 , pages =

  54. [54]

    Borgwardt, Karsten and Ghisu, Elisabetta and. Graph. 2020 , journal =. arxiv , langid =:2011.03854 , primaryclass =

  55. [55]

    Improving

    Bouritsas, Giorgos and Frasca, Fabrizio and Zafeiriou, Stefanos P and Bronstein, Michael , year =. Improving. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

  56. [56]

    Partition and

    Bouritsas, Giorgos and Loukas, Andreas and Karalias, Nikolaos and Bronstein, Michael , year =. Partition and. Advances in

  57. [57]

    2022 , journal =

    Flow Stability for Dynamic Community Detection , author =. 2022 , journal =

  58. [58]

    and Welling, Max , year =

    Brandstetter, Johannes and Worrall, Daniel E. and Welling, Max , year =. Message. International

  59. [59]

    2018 , journal =

    Master Stability Functions Reveal Diffusion-Driven Pattern Formation in Networks , author =. 2018 , journal =. doi:10.1103/PhysRevE.97.032307 , urldate =

  60. [60]

    IEEE Signal Processing Magazine , author =

    Bronstein, Michael M. and Bruna, Joan and LeCun, Yann and Szlam, Arthur and Vandergheynst, Pierre , year =. Geometric Deep Learning: Going beyond. IEEE Signal Processing Magazine , volume =. doi:10.1109/MSP.2017.2693418 , urldate =. arxiv , langid =:1611.08097 , pages =

  61. [61]

    and Bruna, Joan and Cohen, Taco and Veli

    Bronstein, Michael M. and Bruna, Joan and Cohen, Taco and Veli. Geometric. 2021 , journal =. arxiv , langid =:2104.13478 , primaryclass =

  62. [62]

    Spectral

    Bruna, Joan and Zaremba, Wojciech and Szlam, Arthur and LeCun, Yann , year =. Spectral. arXiv:1312.6203 [cs] , eprint =

  63. [63]

    Cai, Chen and Hy, Truong Son and Yu, Rose and Wang, Yusu , year =. On the. Proceedings of the 40th

  64. [64]

    Cai, Chen and Wang, Yusu , year =. A. arXiv:2006.13318 [cs, stat] , eprint =

  65. [65]

    arXiv:2103.00959 [cs, stat] , eprint =

    Cen, Yukuo and Hou, Zhenyu and Wang, Yan and Chen, Qibin and Luo, Yizhen and Yao, Xingcheng and Zeng, Aohan and Guo, Shiguang and Yang, Yang and Zhang, Peng and Dai, Guohao and Wang, Yu and Zhou, Chang and Yang, Hongxia and Tang, Jie , year =. arXiv:2103.00959 [cs, stat] , eprint =

  66. [66]

    , year =

    Chamberlain, Benjamin Paul and Rowbottom, James and Eynard, Davide and Giovanni, Francesco Di and Dong, Xiaowen and Bronstein, Michael M. , year =. Beltrami. Advances in

  67. [67]

    and Bronstein, Michael and Webb, Stefan and Rossi, Emanuele , year =

    Chamberlain, Ben and Rowbottom, James and Gorinova, Maria I. and Bronstein, Michael and Webb, Stefan and Rossi, Emanuele , year =. Proceedings of the 38th

  68. [68]

    and Hansmire, Max , year =

    Chamberlain, Benjamin Paul and Shirobokov, Sergey and Rossi, Emanuele and Frasca, Fabrizio and Markovich, Thomas and Hammerla, Nils Yannick and Bronstein, Michael M. and Hansmire, Max , year =. Graph. The

  69. [69]

    and Hansmire, Max , year =

    Chamberlain, Benjamin Paul and Shirobokov, Sergey and Rossi, Emanuele and Frasca, Fabrizio and Markovich, Thomas and Hammerla, Nils Yannick and Bronstein, Michael M. and Hansmire, Max , year =. Graph. International

  70. [70]

    Structure-

    Chang, Jianlong and Gu, Jie and Wang, Lingfeng and MENG,. Structure-. Advances in. 2018 , volume =

  71. [71]

    Proceedings of the 38th

    Chanpuriya, Sudhanshu and Musco, Cameron and Sotiropoulos, Konstantinos and Tsourakakis, Charalampos , year =. Proceedings of the 38th

  72. [72]

    Bridging the

    Chen, Zhiqian and Chen, Fanglan and Zhang, Lei and Ji, Taoran and Fu, Kaiqun and Zhao, Liang and Chen, Feng and Lu, Chang-Tien , year =. Bridging the. arXiv:2002.11867 [cs, stat] , eprint =

  73. [73]

    Chen, Zhengdao and Chen, Lei and Villar, Soledad and Bruna, Joan , year =. Can. Advances in

  74. [74]

    On the Equivalence between Graph Isomorphism Testing and Function Approximation with

    Chen, Zhengdao and Villar, Soledad and Chen, Lei and Bruna, Joan , year =. On the Equivalence between Graph Isomorphism Testing and Function Approximation with. Advances in

  75. [75]

    FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling

    Chen, Jie and Ma, Tengfei and Xiao, Cao , year =. doi:10.48550/arXiv.1801.10247 , urldate =. 1801.10247 , primaryclass =

  76. [76]

    Proceedings of the

    Chen, Haochen and Perozzi, Bryan and Hu, Yifan and Skiena, Steven , year =. Proceedings of the. doi:10.1609/aaai.v32i1.11849 , urldate =

  77. [77]

    Chen, Kai and Niu, Meng and Chen, Qingcai , year =. A. arXiv:2012.11960 [cs] , eprint =

  78. [78]

    Measuring and

    Chen, Deli and Lin, Yankai and Li, Wei and Li, Peng and Zhou, Jie and Sun, Xu , year =. Measuring and. Proceedings of the AAAI Conference on Artificial Intelligence , volume =

  79. [79]

    Chen, Weiqi and Chen, Ling and Xie, Yu and Cao, Wei and Gao, Yusong and Feng, Xiaojie , year =. Multi-. Proceedings of the 34th. doi:10/gk9mtt , urldate =

  80. [80]

    International

    Chen, Jinsong and Gao, Kaiyuan and Li, Gaichao and He, Kun , year =. International

Showing first 80 references.