FIT-GNN: Faster Inference Time for GNNs that 'FIT' in Memory Using Coarsening

Anirban Dasgupta; Hrriday Ruparel; Kishan Ved; Shubhajit Roy

arxiv: 2410.15001 · v5 · submitted 2024-10-19 · 💻 cs.LG · stat.ML

FIT-GNN: Faster Inference Time for GNNs that 'FIT' in Memory Using Coarsening

Shubhajit Roy , Hrriday Ruparel , Kishan Ved , Anirban Dasgupta This is my paper

Pith reviewed 2026-05-23 19:16 UTC · model grok-4.3

classification 💻 cs.LG stat.ML

keywords graph neural networksgraph coarseninginference timememory efficiencyscalabilitygraph classificationgraph regression

0 comments

The pith

Graph coarsening reduces GNN inference time by orders of magnitude and cuts memory use while keeping accuracy competitive.

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

The paper establishes that graph coarsening, previously used mainly for training, can also speed up the inference stage of GNNs. It introduces two concrete coarsening approaches, Extra Nodes and Cluster Nodes, that replace the original graph with a much smaller one during prediction. Experiments on node classification, graph classification, and graph regression benchmarks show large drops in both inference time per node and overall memory footprint. These reductions allow the same models to run on devices that cannot hold the full graph in memory.

Core claim

Applying the Extra Nodes or Cluster Nodes coarsening strategies produces a reduced graph on which a trained GNN performs inference; this yields orders-of-magnitude faster single-node inference and substantially lower memory consumption for node and graph tasks, while accuracy stays competitive with full-graph baselines across multiple datasets.

What carries the argument

The Extra Nodes and Cluster Nodes coarsening strategies, which construct a smaller proxy graph that the GNN uses at inference time instead of the original input graph.

If this is right

Single-node inference latency falls enough for real-time use on ordinary hardware.
Memory footprint shrinks so that models become runnable on low-resource devices.
The same speed and memory gains apply to both node-level and graph-level prediction tasks.
No retraining of the underlying GNN is required; the coarsened graph is used only at inference.

Where Pith is reading between the lines

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

The approach could be tried on streaming or time-varying graphs where the full structure never fits in memory at once.
Accuracy retention may hinge on how faithfully the coarsening keeps higher-order connectivity patterns such as communities.
The same coarsening steps might be combined with quantization or pruning for even larger efficiency gains.

Load-bearing premise

The coarsened graph still contains enough of the original connectivity and node features for the downstream GNN to produce predictions whose quality is close to what the full graph would yield.

What would settle it

Measurement on a held-out dataset showing that coarsened-graph inference accuracy drops by more than a small fixed margin (for example 5 percentage points) below the full-graph baseline.

Figures

Figures reproduced from arXiv: 2410.15001 by Anirban Dasgupta, Hrriday Ruparel, Kishan Ved, Shubhajit Roy.

**Figure 2.** Figure 2: Methods of appending additional nodes in G1, G2, G3. 4 Methodology In SGGC, a graph G is reduced to G′ using a partition matrix P generated from the coarsening algorithm. Here X′ = P T X. The labels for the coarsened graph are Y ′ = arg max(P T Y ), where Y is the label matrix of G storing the one-hot encoding of the label of each node for the node classification task. Training is carried out on the coarse… view at source ↗

**Figure 3.** Figure 3: GPU memory consumption (log scale) for FIT-GNN (Cluster Node) at different reduction ratios r and Baseline, during inference. 165000 nodes and 4340428 edges. We see up to 100×. Section C elaborates more on the inference time for datasets with a larger number of nodes. Traditional methods fail to infer for very large datasets due to memory constraints. However, our approach can give an inference on the whol… view at source ↗

**Figure 4.** Figure 4: The plot shows the feasibility region (green) corresponding to the coarsening ratio and Number of nodes (In order of Millions) in the graph for a fixed feature dimension of d = 200. In the above equation, we use the notations already defined in Section 4.3. We can also assume that maxk i ni = n α , where α > 1 is some constant. Then we can rewrite the Equation 3 as: T (n) = n 2 d + nd2 − ( n α + ϕmax) 2 d … view at source ↗

read the original abstract

Scalability of Graph Neural Networks (GNNs) remains a significant challenge. To tackle this, methods like coarsening, condensation, and computation trees are used to train on a smaller graph, resulting in faster computation. Nonetheless, prior research has not adequately addressed the computational costs during the inference phase. This paper presents a novel approach to improve the scalability of GNNs by reducing computational burden during the inference phase using graph coarsening. We demonstrate two different methods -- Extra Nodes and Cluster Nodes. Our study extends the application of graph coarsening for graph-level tasks, including graph classification and graph regression. We conduct extensive experiments on multiple benchmark datasets to evaluate the performance of our approach. Our results show that the proposed method achieves orders of magnitude improvements in single-node inference time compared to traditional approaches. Furthermore, it significantly reduces memory consumption for node and graph classification and regression tasks, enabling efficient training and inference on low-resource devices where conventional methods are impractical. Notably, these computational advantages are achieved while maintaining competitive performance relative to baseline models.

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.

Desk Editor's Note private letter to a colleague

The paper adapts coarsening to GNN inference with two node-handling schemes and claims big speedups, but node-level accuracy preservation is the part that needs checking.

read the letter

The main contribution is shifting coarsening from training to inference. They define Extra Nodes and Cluster Nodes schemes to produce a smaller graph that a GNN can run on, then extend the idea to graph classification and regression. That fills a practical gap since most prior coarsening work stopped at training time. The reported outcome is orders-of-magnitude faster single-node inference and lower memory use while accuracy stays competitive on the benchmarks they tried. Those are the concrete pieces that are new and potentially useful for anyone who needs to run GNNs on edge hardware or very large graphs. The soft spot is exactly the one the stress-test flags. Both schemes aggregate or augment nodes, which can erase distinctions that matter for node-level predictions. The abstract asserts competitive performance but gives no indication of how node identity or local structure is preserved, no error bars, and no ablation on coarsening ratio versus accuracy drop. Without those details it is hard to know whether the accuracy claim holds for node tasks or only for graph-level ones. The paper is aimed at practitioners who care about inference cost rather than theorists looking for new GNN layers. A reader who wants reproducible tricks for low-resource deployment would get value from trying the two schemes. It deserves a serious referee to verify the experimental section and see whether the accuracy numbers survive closer inspection.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces FIT-GNN, which applies two graph coarsening strategies (Extra Nodes and Cluster Nodes) to produce reduced graphs on which GNN inference is performed. It claims orders-of-magnitude reductions in single-node inference time and memory consumption for node and graph classification/regression tasks on benchmark datasets, while maintaining competitive accuracy relative to full-graph baselines.

Significance. If the empirical claims are substantiated, the work would address a practical gap in GNN scalability by targeting inference rather than training; this could enable deployment on low-resource devices. Explicit credit is due for the focus on single-node inference timing and the extension of coarsening to graph-level tasks.

major comments (2)

[§3] §3: The Extra Nodes and Cluster Nodes coarsening procedures are not shown to encode original node identities or neighborhoods in the reduced graph; without such mechanisms the claim of competitive accuracy on node-level tasks is at risk because aggregation erases per-node distinctions.
[§4] §4: No error bars, dataset sizes, baseline implementation details, or ablations of coarsening ratio versus accuracy are referenced, so the central empirical claim of orders-of-magnitude speedups with competitive performance cannot be verified from the reported results.

minor comments (1)

The abstract would be strengthened by including at least one concrete quantitative example of reported speedup or accuracy delta.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below, agreeing where revisions are needed to strengthen the manuscript.

read point-by-point responses

Referee: [§3] §3: The Extra Nodes and Cluster Nodes coarsening procedures are not shown to encode original node identities or neighborhoods in the reduced graph; without such mechanisms the claim of competitive accuracy on node-level tasks is at risk because aggregation erases per-node distinctions.

Authors: We thank the referee for this observation. Section 3 presents the two coarsening strategies, with Extra Nodes introducing auxiliary nodes to carry aggregated neighborhood information and Cluster Nodes relying on clustering that retains connectivity patterns. Both include an explicit mapping from the coarsened graph back to the original nodes to enable per-node predictions. However, the current text does not sufficiently highlight this encoding mechanism. We will revise §3 to add a dedicated paragraph and a small illustrative diagram demonstrating how node identities and neighborhoods are preserved through the mapping, ensuring that aggregation does not erase distinctions for node-level tasks. revision: yes
Referee: [§4] §4: No error bars, dataset sizes, baseline implementation details, or ablations of coarsening ratio versus accuracy are referenced, so the central empirical claim of orders-of-magnitude speedups with competitive performance cannot be verified from the reported results.

Authors: We agree that these elements are essential for reproducibility and verification of the empirical claims. The revised manuscript will incorporate error bars (standard deviation across multiple random seeds), explicit dataset sizes for all benchmarks, additional baseline implementation details (library versions, hardware, and hyperparameter settings), and new ablation experiments that vary the coarsening ratio while reporting both accuracy and inference-time metrics. These additions will directly address the verifiability concern. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical claims rest on experiments

full rationale

The paper proposes two coarsening strategies (Extra Nodes, Cluster Nodes) and reports empirical speed/accuracy results on benchmarks. No equations, derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear. Central claims are validated by direct experimental comparison to baselines rather than reducing to definitions or prior author work by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no equations or implementation details, so no free parameters, axioms, or invented entities can be identified with certainty.

pith-pipeline@v0.9.0 · 5729 in / 1081 out tokens · 26250 ms · 2026-05-23T19:16:32.254376+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 5 internal anchors

[1]

Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks

Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’19, page 257–266. Association for Computing Machinery, 2019. ISBN 9781450362016. 2

work page 2019
[2]

Graphsaint: Graph sampling based inductive learning method

Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph sampling based inductive learning method. In International Conference on Learning Representations, 2020. 2

work page 2020
[3]

Stochastic training of graph convolutional networks, 2018

Jianfei Chen and Jun Zhu. Stochastic training of graph convolutional networks, 2018. 2

work page 2018
[4]

FastGCN: Fast learning with graph convolutional networks via importance sampling

Jie Chen, Tengfei Ma, and Cao Xiao. FastGCN: Fast learning with graph convolutional networks via importance sampling. In International Conference on Learning Representations , 2018. 2

work page 2018
[5]

Minimal variance sam- pling with provable guarantees for fast training of graph neural networks

Weilin Cong, Rana Forsati, Mahmut Kandemir, and Mehrdad Mahdavi. Minimal variance sam- pling with provable guarantees for fast training of graph neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’20, page 1393–1403. Association for Computing Machinery, 2020. ISBN 9781450379984. 2

work page 2020
[6]

Layer- dependent importance sampling for training deep and large graph convolutional networks

Difan Zou, Ziniu Hu, Yewen Wang, Song Jiang, Yizhou Sun, and Quanquan Gu. Layer- dependent importance sampling for training deep and large graph convolutional networks . Curran Associates Inc., 2019. 2

work page 2019
[7]

Faster graph embeddings via coarsening

Matthew Fahrbach, Gramoz Goranci, Richard Peng, Sushant Sachdeva, and Chi Wang. Faster graph embeddings via coarsening. CoRR, abs/2007.02817, 2020. 2

work page arXiv 2007
[8]

Scaling up graph neural networks via graph coarsening

Zengfeng Huang, Shengzhong Zhang, Chong Xi, Tang Liu, and Min Zhou. Scaling up graph neural networks via graph coarsening. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , KDD ’21, page 675–684. Association for Computing Machinery, 2021. ISBN 9781450383325. 2, 13

work page 2021
[9]

Featured graph coarsening with similarity guarantees

Manoj Kumar, Anurag Sharma, Shashwat Saxena, and Sandeep Kumar. Featured graph coarsening with similarity guarantees. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors, Proceedings of the 40th International Conference on Machine Learning , volume 202 of Proceedings of Machine Learning Resear...

work page 2023
[10]

Graph condensation for graph neural networks

Wei Jin, Lingxiao Zhao, Shichang Zhang, Yozen Liu, Jiliang Tang, and Neil Shah. Graph condensation for graph neural networks. arXiv preprint arXiv:2110.07580, 2021. 2

work page arXiv 2021
[11]

Fast graph condensation with structure-based neural tangent kernel

Lin Wang, Wenqi Fan, Jiatong Li, Yao Ma, and Qing Li. Fast graph condensation with structure-based neural tangent kernel. In Proceedings of the ACM Web Conference 2024, pages 4439–4448, 2024. 2

work page 2024
[12]

Bonsai: Gradient-free graph condensation for node classification

Mridul Gupta, Samyak Jain, Vansh Ramani, Hariprasad Kodamana, and Sayan Ranu. Bonsai: Gradient-free graph condensation for node classification. In The Thirteenth International Conference on Learning Representations , 2025. URL https://openreview.net/forum? id=5x88lQ2MsH. 2

work page 2025
[13]

SUGAR: efficient subgraph- level training via resource-aware graph partitioning

Zihui Xue, Yuedong Yang, Mengtian Yang, and Radu Marculescu. SUGAR: efficient subgraph- level training via resource-aware graph partitioning. CoRR, abs/2202.00075, 2022. 2, 4

work page arXiv 2022
[14]

Condensing graphs via one-step gradient matching

Wei Jin, Xianfeng Tang, Haoming Jiang, Zheng Li, Danqing Zhang, Jiliang Tang, and Bing Yin. Condensing graphs via one-step gradient matching. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD ’22, page 720–730. ACM, August 2022. doi: 10.1145/3534678.3539429. URL http://dx.doi.org/10.1145/ 3534678.3539429. 2

work page doi:10.1145/3534678.3539429 2022
[15]

Kernel ridge regression-based graph dataset distillation

Zhe Xu, Yuzhong Chen, Menghai Pan, Huiyuan Chen, Mahashweta Das, Hao Yang, and Hanghang Tong. Kernel ridge regression-based graph dataset distillation. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD ’23, page 2850–2861, New York, NY , USA, 2023. Association for Computing Machinery. ISBN 9798400701030. doi: 10...

work page doi:10.1145/3580305.3599398 2023
[16]

Kipf and Max Welling

Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations , 2017. 3 10 FIT-GNN: Faster Inference Time for GNNs that ‘FIT’ in Memory Using Coarsening

work page 2017
[17]

Graph reduction with spectral and cut guarantees, 2018

Andreas Loukas. Graph reduction with spectral and cut guarantees, 2018. 3

work page 2018
[18]

Scalable and effective implicit graph neural networks on large graphs

Juncheng Liu, Bryan Hooi, Kenji Kawaguchi, Yiwei Wang, Chaosheng Dong, and Xiaokui Xiao. Scalable and effective implicit graph neural networks on large graphs. In The Twelfth International Conference on Learning Representations , 2024. 4

work page 2024
[19]

Graph Attention Networks

Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks, 2018. URL https://arxiv.org/abs/1710.10903. 5

work page internal anchor Pith review Pith/arXiv arXiv 2018
[20]

Multi-scale attributed node embedding,

Benedek Rozemberczki, Carl Allen, and Rik Sarkar. Multi-scale attributed node embedding,

work page
[21]

Automating the construction of internet portals with machine learning

Andrew Kachites McCallum, Kamal Nigam, Jason Rennie, and Kristie Seymore. Automating the construction of internet portals with machine learning. Information Retrieval, 3(2):127–163, Jul 2000. ISSN 1573-7659. 7

work page 2000
[22]

Lee Giles, Kurt D

C. Lee Giles, Kurt D. Bollacker, and Steve Lawrence. Citeseer: an automatic citation indexing system. In Proceedings of the Third ACM Conference on Digital Libraries , DL ’98, page 89–98. Association for Computing Machinery, 1998. ISBN 0897919653. 7

work page 1998
[23]

Collective classification in network data

Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi- Rad. Collective classification in network data. AI Magazine, 29(3):93, Sep. 2008. 7

work page 2008
[24]

Pitfalls of Graph Neural Network Evaluation

Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan Günnemann. Pitfalls of graph neural network evaluation. CoRR, abs/1811.05868, 2018. 7

work page internal anchor Pith review Pith/arXiv arXiv 2018
[25]

Arnetminer: extraction and mining of academic social networks

Jie Tang, Jing Zhang, Limin Yao, Juanzi Li, Li Zhang, and Zhong Su. Arnetminer: extraction and mining of academic social networks. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’08, page 990–998. Association for Computing Machinery, 2008. ISBN 9781605581934. 7

work page 2008
[26]

Open graph benchmark: datasets for machine learning on graphs

Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: datasets for machine learning on graphs. In Proceedings of the 34th International Conference on Neural Information Processing Systems , NIPS ’20. Curran Associates Inc., 2020. ISBN 9781713829546. 7

work page 2020
[27]

MoleculeNet: A Benchmark for Molecular Machine Learning

Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, and Vijay S. Pande. Moleculenet: A benchmark for molecular machine learning. CoRR, abs/1703.00564, 2017. 7

work page internal anchor Pith review Pith/arXiv arXiv 2017
[28]

Automatic chemical design using a data-driven continuous representation of molecules

Rafael Gómez-Bombarelli, David Duvenaud, José Miguel Hernández-Lobato, Jorge Aguilera- Iparraguirre, Timothy D. Hirzel, Ryan P. Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven continuous representation of molecules. CoRR, abs/1610.02415,

work page internal anchor Pith review Pith/arXiv arXiv
[29]

Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann

Christopher Morris, Nils M. Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann. Tudataset: A collection of benchmark datasets for learning with graphs. In ICML 2020 Workshop on Graph Representation Learning and Beyond (GRL+ 2020) , 2020. 7

work page 2020
[30]

Cohen, and Ruslan Salakhutdinov

Zhilin Yang, William W. Cohen, and Ruslan Salakhutdinov. Revisiting semi-supervised learning with graph embeddings. In Proceedings of the 33rd International Conference on International Conference on Machine Learning - V olume 48, ICML’16, page 40–48. JMLR.org, 2016. 7

work page 2016
[31]

Neural Message Passing for Quantum Chemistry

Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. Neural message passing for quantum chemistry. CoRR, abs/1704.01212, 2017. 7 11 FIT-GNN: Faster Inference Time for GNNs that ‘FIT’ in Memory Using Coarsening Appendix A More on Extra Nodes and Cluster Nodes A.1 Extra Nodes Lemma 4.1. Models with 1 layer of GNN cannot ...

work page internal anchor Pith review Pith/arXiv arXiv 2017

[1] [1]

Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks

Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’19, page 257–266. Association for Computing Machinery, 2019. ISBN 9781450362016. 2

work page 2019

[2] [2]

Graphsaint: Graph sampling based inductive learning method

Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph sampling based inductive learning method. In International Conference on Learning Representations, 2020. 2

work page 2020

[3] [3]

Stochastic training of graph convolutional networks, 2018

Jianfei Chen and Jun Zhu. Stochastic training of graph convolutional networks, 2018. 2

work page 2018

[4] [4]

FastGCN: Fast learning with graph convolutional networks via importance sampling

Jie Chen, Tengfei Ma, and Cao Xiao. FastGCN: Fast learning with graph convolutional networks via importance sampling. In International Conference on Learning Representations , 2018. 2

work page 2018

[5] [5]

Minimal variance sam- pling with provable guarantees for fast training of graph neural networks

Weilin Cong, Rana Forsati, Mahmut Kandemir, and Mehrdad Mahdavi. Minimal variance sam- pling with provable guarantees for fast training of graph neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’20, page 1393–1403. Association for Computing Machinery, 2020. ISBN 9781450379984. 2

work page 2020

[6] [6]

Layer- dependent importance sampling for training deep and large graph convolutional networks

Difan Zou, Ziniu Hu, Yewen Wang, Song Jiang, Yizhou Sun, and Quanquan Gu. Layer- dependent importance sampling for training deep and large graph convolutional networks . Curran Associates Inc., 2019. 2

work page 2019

[7] [7]

Faster graph embeddings via coarsening

Matthew Fahrbach, Gramoz Goranci, Richard Peng, Sushant Sachdeva, and Chi Wang. Faster graph embeddings via coarsening. CoRR, abs/2007.02817, 2020. 2

work page arXiv 2007

[8] [8]

Scaling up graph neural networks via graph coarsening

Zengfeng Huang, Shengzhong Zhang, Chong Xi, Tang Liu, and Min Zhou. Scaling up graph neural networks via graph coarsening. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , KDD ’21, page 675–684. Association for Computing Machinery, 2021. ISBN 9781450383325. 2, 13

work page 2021

[9] [9]

Featured graph coarsening with similarity guarantees

Manoj Kumar, Anurag Sharma, Shashwat Saxena, and Sandeep Kumar. Featured graph coarsening with similarity guarantees. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors, Proceedings of the 40th International Conference on Machine Learning , volume 202 of Proceedings of Machine Learning Resear...

work page 2023

[10] [10]

Graph condensation for graph neural networks

Wei Jin, Lingxiao Zhao, Shichang Zhang, Yozen Liu, Jiliang Tang, and Neil Shah. Graph condensation for graph neural networks. arXiv preprint arXiv:2110.07580, 2021. 2

work page arXiv 2021

[11] [11]

Fast graph condensation with structure-based neural tangent kernel

Lin Wang, Wenqi Fan, Jiatong Li, Yao Ma, and Qing Li. Fast graph condensation with structure-based neural tangent kernel. In Proceedings of the ACM Web Conference 2024, pages 4439–4448, 2024. 2

work page 2024

[12] [12]

Bonsai: Gradient-free graph condensation for node classification

Mridul Gupta, Samyak Jain, Vansh Ramani, Hariprasad Kodamana, and Sayan Ranu. Bonsai: Gradient-free graph condensation for node classification. In The Thirteenth International Conference on Learning Representations , 2025. URL https://openreview.net/forum? id=5x88lQ2MsH. 2

work page 2025

[13] [13]

SUGAR: efficient subgraph- level training via resource-aware graph partitioning

Zihui Xue, Yuedong Yang, Mengtian Yang, and Radu Marculescu. SUGAR: efficient subgraph- level training via resource-aware graph partitioning. CoRR, abs/2202.00075, 2022. 2, 4

work page arXiv 2022

[14] [14]

Condensing graphs via one-step gradient matching

Wei Jin, Xianfeng Tang, Haoming Jiang, Zheng Li, Danqing Zhang, Jiliang Tang, and Bing Yin. Condensing graphs via one-step gradient matching. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD ’22, page 720–730. ACM, August 2022. doi: 10.1145/3534678.3539429. URL http://dx.doi.org/10.1145/ 3534678.3539429. 2

work page doi:10.1145/3534678.3539429 2022

[15] [15]

Kernel ridge regression-based graph dataset distillation

Zhe Xu, Yuzhong Chen, Menghai Pan, Huiyuan Chen, Mahashweta Das, Hao Yang, and Hanghang Tong. Kernel ridge regression-based graph dataset distillation. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD ’23, page 2850–2861, New York, NY , USA, 2023. Association for Computing Machinery. ISBN 9798400701030. doi: 10...

work page doi:10.1145/3580305.3599398 2023

[16] [16]

Kipf and Max Welling

Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations , 2017. 3 10 FIT-GNN: Faster Inference Time for GNNs that ‘FIT’ in Memory Using Coarsening

work page 2017

[17] [17]

Graph reduction with spectral and cut guarantees, 2018

Andreas Loukas. Graph reduction with spectral and cut guarantees, 2018. 3

work page 2018

[18] [18]

Scalable and effective implicit graph neural networks on large graphs

Juncheng Liu, Bryan Hooi, Kenji Kawaguchi, Yiwei Wang, Chaosheng Dong, and Xiaokui Xiao. Scalable and effective implicit graph neural networks on large graphs. In The Twelfth International Conference on Learning Representations , 2024. 4

work page 2024

[19] [19]

Graph Attention Networks

Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks, 2018. URL https://arxiv.org/abs/1710.10903. 5

work page internal anchor Pith review Pith/arXiv arXiv 2018

[20] [20]

Multi-scale attributed node embedding,

Benedek Rozemberczki, Carl Allen, and Rik Sarkar. Multi-scale attributed node embedding,

work page

[21] [21]

Automating the construction of internet portals with machine learning

Andrew Kachites McCallum, Kamal Nigam, Jason Rennie, and Kristie Seymore. Automating the construction of internet portals with machine learning. Information Retrieval, 3(2):127–163, Jul 2000. ISSN 1573-7659. 7

work page 2000

[22] [22]

Lee Giles, Kurt D

C. Lee Giles, Kurt D. Bollacker, and Steve Lawrence. Citeseer: an automatic citation indexing system. In Proceedings of the Third ACM Conference on Digital Libraries , DL ’98, page 89–98. Association for Computing Machinery, 1998. ISBN 0897919653. 7

work page 1998

[23] [23]

Collective classification in network data

Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi- Rad. Collective classification in network data. AI Magazine, 29(3):93, Sep. 2008. 7

work page 2008

[24] [24]

Pitfalls of Graph Neural Network Evaluation

Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan Günnemann. Pitfalls of graph neural network evaluation. CoRR, abs/1811.05868, 2018. 7

work page internal anchor Pith review Pith/arXiv arXiv 2018

[25] [25]

Arnetminer: extraction and mining of academic social networks

Jie Tang, Jing Zhang, Limin Yao, Juanzi Li, Li Zhang, and Zhong Su. Arnetminer: extraction and mining of academic social networks. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’08, page 990–998. Association for Computing Machinery, 2008. ISBN 9781605581934. 7

work page 2008

[26] [26]

Open graph benchmark: datasets for machine learning on graphs

Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: datasets for machine learning on graphs. In Proceedings of the 34th International Conference on Neural Information Processing Systems , NIPS ’20. Curran Associates Inc., 2020. ISBN 9781713829546. 7

work page 2020

[27] [27]

MoleculeNet: A Benchmark for Molecular Machine Learning

Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, and Vijay S. Pande. Moleculenet: A benchmark for molecular machine learning. CoRR, abs/1703.00564, 2017. 7

work page internal anchor Pith review Pith/arXiv arXiv 2017

[28] [28]

Automatic chemical design using a data-driven continuous representation of molecules

Rafael Gómez-Bombarelli, David Duvenaud, José Miguel Hernández-Lobato, Jorge Aguilera- Iparraguirre, Timothy D. Hirzel, Ryan P. Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven continuous representation of molecules. CoRR, abs/1610.02415,

work page internal anchor Pith review Pith/arXiv arXiv

[29] [29]

Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann

Christopher Morris, Nils M. Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann. Tudataset: A collection of benchmark datasets for learning with graphs. In ICML 2020 Workshop on Graph Representation Learning and Beyond (GRL+ 2020) , 2020. 7

work page 2020

[30] [30]

Cohen, and Ruslan Salakhutdinov

Zhilin Yang, William W. Cohen, and Ruslan Salakhutdinov. Revisiting semi-supervised learning with graph embeddings. In Proceedings of the 33rd International Conference on International Conference on Machine Learning - V olume 48, ICML’16, page 40–48. JMLR.org, 2016. 7

work page 2016

[31] [31]

Neural Message Passing for Quantum Chemistry

Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. Neural message passing for quantum chemistry. CoRR, abs/1704.01212, 2017. 7 11 FIT-GNN: Faster Inference Time for GNNs that ‘FIT’ in Memory Using Coarsening Appendix A More on Extra Nodes and Cluster Nodes A.1 Extra Nodes Lemma 4.1. Models with 1 layer of GNN cannot ...

work page internal anchor Pith review Pith/arXiv arXiv 2017