Pith. sign in

REVIEW 3 major objections 6 minor 80 references

ReInc: Scaling Training of Dynamic Graph Neural Networks

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read ReInc claims that distributed training of dynamic graph neural networks can be made communication-free in the forward pass and up to 17.7x faster than prior systems by reusing intermediate aggregations and placing graph snapshots as…

desk verdict Correct incremental-aggregation core, plausible speedups, but the headline zero-communication claim rests on an unevaluated snapshot-overlap branch the paper itself only mentions in passing. read the letter →

arxiv 2501.15348 v1 pith:L3UES7JC submitted 2025-01-25 cs.LG cs.DC

Mingyu Guan , Saumia Singhal , Taesoo Kim , Anand Padmanabha Iyer This is my paper · ORCID
classification cs.LGcs.DC
keywords dynamicgraphneuralnetworksincrementalaggregationdistributedtrainingsnapshotreusetwo-levelcachingmini-batchtrafficforecastingpartitioning
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 proposes REINC, a training system for dynamic graph neural networks (DGNNs) that combine a GNN with an RNN. Its claim is that three reuse opportunities—across RNN gates, across overlapping sliding-window sequences, and between decoder and encoder under teacher forcing—plus incremental aggregation over delta graphs, a two-level DGNN-aware cache, and a consecutive-block snapshot placement that keeps each training sequence local, make distributed DGNN training faster by roughly an order of magnitude. On four DGNN architectures and four large graphs it reports 2.9–12.8× speedups over DynaGraph and 2.8–17.7× over ESDGNN, with the same test MAE in a correctness comparison. The broader importance is that scaling DGNNs to large dynamic graphs has been blocked by communication and redundant computation; if REINC is right, both bottlenecks can be removed without sacrificing accuracy.

What carries the argument

The load-bearing identity is incremental aggregation over delta graphs: $\mathrm{Agg}_t = \mathrm{Agg}_{t-1} - F_{t-1} \ast_{\text{aggr}} G^-_t + F_t \ast_{\text{aggr}} G^+_t$, where $G^-_t$ and $G^+_t$ collect edge deletions and insertions between consecutive snapshots and feature changes are rewritten as deletions plus insertions. This turns recomputing each snapshot from scratch into a small correction, which is effective because real-world dynamic graphs change slowly relative to their size. The accompanying machinery is a two-level cache store with a DGNN-aware priority score based on future access count, imminence, and size, plus consecutive-block snapshot placement with seq-first mini-batch iteration so that cached aggregations are reused while sequences stay local. The identity carries the computational savings; the placement carries the communication savings.

What would settle it

Run the same four DGNNs with a sequence length larger than the per-machine snapshot budget on a graph whose snapshots exceed one machine's host memory, and compare per-epoch time and network bytes against the paper's zero-communication claim; if remote snapshot fetch dominates, the claimed scaling does not hold. Alternatively, train a max()-aggregation model on a graph with frequent edge deletions, where the paper's own fallback to from-scratch aggregation should eliminate the incremental speedup.

Watch

Extended reading notes

Core claim

REINC's central claim is that the execution of a DGNN, whether the GNN and RNN are stacked or integrated into a GraphRNN, can be made communication-free in the forward pass by placing consecutive snapshots as blocks on machines: because time dependencies exist only within a training sequence and sequences are independent, each machine holds the full sequence of snapshots its mini-batch needs, so no remote feature pulls or intermediate redistribution occur. On top of this placement, REINC avoids recomputation by caching and reusing aggregations and by computing each snapshot's aggregation from the previous one using only the changed edges, with feature changes encoded as edge deletions and insertions. The paper supports the claim with experiments showing 2.9–12.8× and 2.8–17.7× epoch-time speedups over DynaGraph and ESDGNN respectively across GCRN-M1, CD-GCN, GCRN-M2, and T-GCN on four large graphs, and identical test MAE on METR-LA-LARGE with 2.9× and 8.1× speedups.

Load-bearing premise

The load-bearing premise is that each machine's assigned training sequences are fully local after snapshot placement; if a machine lacks memory to hold overlapping snapshots, it must fetch remote snapshots, which would replace the headline zero communication with network traffic.

Editorial extensions

If this is right

  • REINC's reuse and incremental aggregation make integrated GraphRNN architectures trainable at scale, closing the gap that previous systems left for stacked-only or integrated-only optimization.
  • Longer sequences and larger feature or hidden dimensions no longer create proportional communication and recomputation overhead, so DGNN training can scale to more history and richer features.
  • The seq-first mini-batch order enables near-complete cache reuse at modest cache sizes, reducing GPU memory pressure during training.
  • Because the correctness run matches prior test MAE, the distributed strategy and optimizations can be adopted without changing model accuracy.

Reading between the lines

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

  • An implication the paper leaves implicit is that the zero-communication forward pass is contingent on sequence locality; for sequences longer than a machine's snapshot block, the fallback of remote snapshot fetch would reintroduce network traffic and should be measured.
  • The same delta-graph incremental aggregation could be applied to streaming or continuous-time GNN training, where changes arrive as edge events rather than discrete snapshots, but attention-style weighted aggregations would need re-aggregation after weight updates.
  • If change ratios are high in a deployment, the incremental speedup shrinks toward the from-scratch baseline; the paper's own fallback threshold makes the benefit workload-dependent, so the reported speedups generalize best to slowly evolving graphs.
  • The seq-first mini-batch strategy is a general scheduling idea for any sequence-of-snapshots workload: iterating over time before sampling nodes maximizes reuse of cached intermediate results.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper presents ReInc, a system for distributed training of discrete dynamic graph neural networks (DGNNs). It identifies three reuse opportunities (across RNN gates, across overlapping sliding-window sequences, and between encoder/decoder teacher-forcing inputs), proposes incremental aggregation through delta graphs, introduces a two-level cache with a DGNN-aware eviction policy, and proposes a consecutive-block snapshot placement with sequence-first mini-batching that is claimed to eliminate remote feature access and intermediate-result redistribution. The evaluation compares ReInc with DynaGraph and ESDGNN on four DGNN architectures and four large graph datasets, reporting speedups of 2.9--12.8x and 2.8--17.7x respectively, and identical test MAE on a correctness check.

Significance. If the claims hold, ReInc addresses a real bottleneck in scaling DGNN training: the combination of graph structure and temporal dependencies causes redundant computation and communication in existing systems. The incremental aggregation identity in Eq. (2) is correct for sum aggregation, and the reuse-based computation preserves exact aggregate values, so the core optimization is mathematically sound. The paper also provides a broad empirical comparison across multiple architectures and baselines, and it explicitly checks correctness against an independent baseline with matching MAE. The main limitations are that the headline zero-communication distributed claim is conditional on an unmeasured memory-replication branch, the incremental aggregation treatment is incomplete for mean (and partly for max/min), and the two largest datasets are synthetic dynamizations of static graphs. These issues are fixable but currently prevent the central claims from being accepted as stated.

major comments (3)
  1. [§3.2, Eq. (2)] Equation (2) is presented as the general incremental aggregation rule, but it is only valid for sum aggregation. For mean aggregation, Agg_t is defined in Eq. (1) as the mean of neighbor features, yet subtracting and adding raw neighbor features does not account for the change in degree. Concretely, if at t-1 a node has neighbor features {1,3} with mean 2, and at t the edge to feature 3 is deleted and an edge to feature 5 is inserted, Eq. (2) gives 2 - 3 + 5 = 4, whereas the true new mean is (1+5)/2 = 3. If ReInc internally stores unnormalized sums and divides by degree only at consumption, that design must be stated and Eq. (2) must be rewritten accordingly; otherwise the claim that mean() is a supported built-in incremental aggregation is unsupported.
  2. [§3.4.1 and Table 1] The '0 GB communication' entry in Table 1 and the abstract's claim of eliminating remote feature access apply only to the overlapped snapshot placement branch of §3.4.1. The paper does not report whether the distributed experiments used overlapped placement, how much additional host memory the overlap required, or the network bytes and time incurred when the fallback branch ('retrieves remote snapshots from other machines during training as needed') is used. For the evaluation configuration with T=100 snapshots, M=8 machines, and L=8, the overlap branch requires roughly (M-1)(L-1)=49 additional snapshot copies across the cluster, and on OGB-Papers a single snapshot already has 1.6B edges and 111M nodes. This is load-bearing because the central distributed claim is that ReInc eliminates communication; please report branch selection, memory overhead, and fallback communication cost, or qualify the abstract and Table 1.
  3. [§5, Experimental Setup] The two largest datasets, OGB-Products and OGB-Papers, are static graphs converted into dynamic ones by randomly modifying edges and features with change ratios drawn uniformly from 0% to 100%, and the traffic datasets are scaled by replication. The headline speedups are therefore measured on synthetic dynamism rather than on real large dynamic graphs, and random rewiring at 100% change is not the 'slowly changing' regime that motivates incremental aggregation. This limits the external validity of the central performance claim. The paper acknowledges the scarcity of public large DGNN datasets, but the abstract's phrase 'real-world graph datasets' overstates the evidence; please add at least one real large dynamic dataset or a sensitivity analysis that varies the structure of changes.
minor comments (6)
  1. [§3.2] The user-defined threshold for falling back to from-scratch aggregation when the change ratio is high is never given a default value, and the experiments in Fig. 13 that vary change ratio do not report when the fallback was triggered.
  2. [§3.3] The text says 'cached aggregations in the global cache can be assessed across layers'; 'assessed' should be 'accessed'.
  3. [§3.3.3] Equation (3), Priority = F(Agg)/S(Agg) - I(Agg), mixes dimensionless future access count, size, and timestep-based imminence without specifying normalization, and the claim of equal weights is not tested via an ablation.
  4. [§5.6] Correctness is shown only for METR-LA-LARGE; the statement that convergence curves on all other datasets 'align consistently' with the baselines is not accompanied by a figure or quantitative comparison.
  5. [Table 1] Table 1 reports epoch communication volume and time but does not describe the dataset, model, sequence length, or hardware configuration used to produce those numbers; a caption or a pointer to the experimental setup is needed.
  6. [Title/Abstract] The title uses 'ReInc' while the body and abstract consistently use 'REINC'; please unify the notation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: REINC's gains come from algebraic reuse and measured comparisons; the sole self-citation (DynaGraph) is a re-implemented baseline, not a load-bearing premise.

full rationale

REINC's derivation chain is not circular. The incremental-aggregation identity (Eq. 2) is an algebraic rewriting of the from-scratch aggregation (Eq. 1) for additive aggregators, with explicit fallbacks to from-scratch computation for max/min, attention weights, and high-change-ratio snapshots; it is not a fitted parameter renamed as a prediction. The reuse, caching, and seq-first mini-batching claims are evaluated as measured system optimizations against DGL-based execution, LRU/LFU policies, and node-first mini-batching, respectively, rather than being assumed by construction. The zero-communication claim is conditional on the overlapped snapshot placement branch of Section 3.4.1; the paper does not quantify the extra memory required for that branch or the network traffic of the fallback branch. That is an unmeasured assumption that weakens the headline distributed-communication claim, but it is a scope/robustness concern, not circularity, because the claim is not used as evidence for itself. The only self-citation is DynaGraph [15], which shares authors with REINC; it appears as a re-implemented baseline, and the paper states that REINC's mini-batch techniques were ported to the baseline, reducing rather than manufacturing the speedup. Correctness is checked against baseline MAE curves on the same prediction task, and no uniqueness theorem or prior self-cited result is invoked to forbid alternative designs. A score of 1 reflects a minor non-load-bearing self-citation, while the central computational and distributed design has independent empirical content.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on two hand-chosen implementation parameters (change-ratio fallback threshold, equal cache-scoring weights) and on standard systems assumptions about discrete snapshots, independent sequences, and sampling. No new physical or mathematical entities are introduced. The incremental aggregation identity is standard linear algebra under the stated domain assumptions.

free parameters (2)
  • change ratio fallback threshold = not specified
    Determines when incremental aggregation switches to from-scratch aggregation; the paper does not report the value used in experiments, so the incremental speedups cannot be reproduced without it.
  • cache scoring weights = equal weights for F/S and imminence
    The cache priority in Eq. 3 combines F/S and imminence with equal weights, chosen by hand without sensitivity analysis; the reported cache hit rates and throughput depend on this choice.
assumptions (4)
  • standard math Linearity of sum and mean aggregation
    Eq. 2 assumes additive aggregation so that Agg_t = Agg_{t-1} minus old contributions plus new contributions; this is exact for sum and mean on unweighted graphs, and the paper explicitly falls back to from-scratch for max and min on deletions.
  • domain assumption Discrete snapshot representation with constant node set within a sequence
    The paper focuses on discrete dynamic graphs and assumes constant matrix sizes within a sequence; node additions and deletions are handled by padding to a maximum node count, stated in Section 3.2.
  • domain assumption Sequences are independent training samples
    The consecutive-block placement and seq-first minibatching rely on the claim that sliding-window sequences are independent samples, stated in Sections 3.4.1 and 3.4.3.
  • domain assumption Neighborhood sampling preserves task correctness
    Experiments use [25,10] sampling and claim correctness; sampling changes the computation graph compared with full neighborhoods, and the paper does not compare against full-graph training for the largest graphs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ReInc: Scaling Training of Dynamic Graph Neural Networks." pith.science (2026). https://pith.science/paper/L3UES7JC

@misc{pith2026250115348,
  author       = {Pith},
  title        = {Pith review of: ReInc: Scaling Training of Dynamic Graph Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L3UES7JC}},
  note         = {Machine review of arXiv:2501.15348}
}
read the original abstract

Dynamic Graph Neural Networks (DGNNs) have gained widespread attention due to their applicability in diverse domains such as traffic network prediction, epidemiological forecasting, and social network analysis. In this paper, we present ReInc, a system designed to enable efficient and scalable training of DGNNs on large-scale graphs. ReInc introduces key innovations that capitalize on the unique combination of Graph Neural Networks (GNNs) and Recurrent Neural Networks (RNNs) inherent in DGNNs. By reusing intermediate results and incrementally computing aggregations across consecutive graph snapshots, ReInc significantly enhances computational efficiency. To support these optimizations, ReInc incorporates a novel two-level caching mechanism with a specialized caching policy aligned to the DGNN execution workflow. Additionally, ReInc addresses the challenges of managing structural and temporal dependencies in dynamic graphs through a new distributed training strategy. This approach eliminates communication overheads associated with accessing remote features and redistributing intermediate results. Experimental results demonstrate that ReInc achieves up to an order of magnitude speedup compared to state-of-the-art frameworks, tested across various dynamic GNN architectures and real-world graph datasets.

Figures

Figures reproduced from arXiv: 2501.15348 by the authors.

Figure 1
Figure 1. (a) A stacked DGNN where an RNN takes the output [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Sliding window mechanism to generate sequences in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Different graph partitioning schemes for a dynamic graph with [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Overview of distributed DGNN execution in R [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Overall, REINC’s speedups range up to 12.8× and 17.7× compared to DynaGraph and ESDGNN, respectively. store using the model GCRN-M2 since an integrated model covers all resuing opportunities so that we can demonstrate the effectiveness of REINC’s techniques. 5.1 Overal…
Figure 8
Figure 8. Figure 8: Without sampling, REINC maintains its benefits while baselines struggle to handle larger graphs. 0.0× 1.0× 2.0× 3.0× 4.0× 5.0× GCRN-M1 CD-GCN GCRN-M2 T-GCN total speedup incremental agg. reuse across gates reuse across seq. reuse in teaching Normalized Speedup [PITH_F…
Figure 9
Figure 9. Figure 9: Impact of REINC ’s reusable optimizations. high memory overhead when pulling computational graphs, and ESDGNN requires an extremely long training time for integrated DGNNs. For METR-LA-LARGE, the epoch times remain similar with or without sampling. However, for OGB￾Pro…
Figure 11
Figure 11. Figure 11: REINC’s cache policy significantly outperforms LRU and LFU for DGNN training. 0 20 40 60 80 100 10 20 30 40 50 60 70 80 90100 10 30 50 70 0x 5x 10x 15x 20x 25x 30x node-first seq-first Cache Hit Rate (%) Cache Size to Data Size (%) node-first seq-first Normalized Thro…
Figure 12
Figure 12. Figure 12: REINC’s seq-first strategy increases cache hit rate and throughput, compared with the standard node-first strategy. to the standard node-first approach. With a small cache-to￾data size (10% to 20%), both strategies achieve similar cache hit rates. However, as cache si…
Figure 14
Figure 14. Figure 14: Correctness of REINC ’s optimizations and dis￾tributed training strategy. DynaGraph, and ESDGNN. As shown in fig. 14, REINC can achieve the same MAE as DynaGraph and ESDGNN, thus ensuring its correctness while being able to gain 2.9× and 8.1× speedup, compared to Dyna…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 66 canonical work pages

  1. [1]

    https://dot.ca.gov/programs/ traffic-operations/mpr/pems-source

    Performance Measurement System (PeMS) Data Source. https://dot.ca.gov/programs/ traffic-operations/mpr/pems-source

    work page

  2. [2]

    Tesseract: distributed, general graph pattern mining on evolving graphs

    Laurent Bindschaedler, Jasmina Malicevic, Baptiste Lep- ers, Ashvin Goel, and Willy Zwaenepoel. Tesseract: distributed, general graph pattern mining on evolving graphs. In Proceedings of the Sixteenth European Conference on Computer Systems, EuroSys ’21, page 458–473, New York, NY , USA, 2021. Association for Computing Machinery

    work page 2021

  3. [3]

    Structural temporal graph neural networks for anomaly detection in dynamic graphs

    Lei Cai, Zhengzhang Chen, Chen Luo, Jiaping Gui, Jingchao Ni, Ding Li, and Haifeng Chen. Structural temporal graph neural networks for anomaly detection in dynamic graphs. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, CIKM ’21, page 3747–3756, New York, NY , USA, 2021. Association for Computing Machinery

    work page 2021

  4. [4]

    Chakaravarthy, Shivmaran S

    Venkatesan T. Chakaravarthy, Shivmaran S. Pandian, Saurabh Raje, Yogish Sabharwal, Toyotaro Suzumura, and Shashanka Ubaru. Efficient scaling of dynamic graph neural networks. In Proceedings of the Inter- national Conference for High Performance Computing, Networking, Storage and Analysis, SC ’21, New York, NY , USA, 2021. Association for Computing Machinery

    work page 2021

  5. [5]

    Gc-lstm: Graph convolution embedded lstm for dynamic link pre- diction, 2021

    Jinyin Chen, Xueke Wang, and Xuanheng Xu. Gc-lstm: Graph convolution embedded lstm for dynamic link pre- diction, 2021

    work page 2021

  6. [6]

    Powerlyra: Differentiated graph computation and parti- tioning on skewed graphs

    Rong Chen, Jiaxin Shi, Yanzhe Chen, and Haibo Chen. Powerlyra: Differentiated graph computation and parti- tioning on skewed graphs. In Proceedings of the Tenth European Conference on Computer Systems, EuroSys ’15, New York, NY , USA, 2015. Association for Com- puting Machinery

    work page 2015

  7. [7]

    Improving WWW proxies per- formance with greedy-dual-size-frequency caching pol- icy

    Ludmila Cherkasova. Improving WWW proxies per- formance with greedy-dual-size-frequency caching pol- icy. Hewlett-Packard Laboratories Palo Alto, CA, USA, 1998

    work page 1998

  8. [8]

    One trillion edges: Graph processing at facebook-scale

    Avery Ching, Sergey Edunov, Maja Kabiljo, Dionysios Logothetis, and Sambavi Muthukrishnan. One trillion edges: Graph processing at facebook-scale. Proc. VLDB Endow., 8(12):1804–1815, August 2015

    work page 2015

Show all 80 references
  1. [9]

    Learning phrase representations using RNN encoder-decoder for statistical machine translation

    Kyunghyun Cho, Bart van Merrienboer, Çaglar Gülçehre, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using RNN encoder-decoder for statistical machine translation. CoRR, abs/1406.1078, 2014. 13

    2014 arXiv

  2. [10]

    Convolutional neural networks on graphs with fast localized spectral filtering

    Michaël Defferrard, Xavier Bresson, and Pierre Van- dergheynst. Convolutional neural networks on graphs with fast localized spectral filtering. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Sys- tems, volume 29...

    2016

  3. [11]

    Matthias Fey and Jan E. Lenssen. Fast graph repre- sentation learning with PyTorch Geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019

    2019

  4. [12]

    P3: Dis- tributed deep graph learning at scale

    Swapnil Gandhi and Anand Padmanabha Iyer. P3: Dis- tributed deep graph learning at scale. In 15th USENIX Symposium on Operating Systems Design and Imple- mentation (OSDI 21), pages 551–568. USENIX Associ- ation, July 2021

    2021

  5. [13]

    Automating incremental graph processing with flexible memoization

    Shufeng Gong, Chao Tian, Qiang Yin, Wenyuan Yu, Yanfeng Zhang, Liang Geng, Song Yu, Ge Yu, and Jin- gren Zhou. Automating incremental graph processing with flexible memoization. Proceedings of the VLDB Endowment, 14(9):1613–1625, 2021

    2021

  6. [14]

    Powergraph: Distributed graph-parallel computation on natural graphs

    Joseph E Gonzalez, Yucheng Low, Haijie Gu, Danny Bickson, and Carlos Guestrin. Powergraph: Distributed graph-parallel computation on natural graphs. In Pre- sented as part of the 10th {USENIX} Symposium on Op- erating Systems Design and Implementation ({OSDI} 12), pages 17–30, 2012

    2012

  7. [15]

    Dynagraph: Dynamic graph neural networks at scale

    Mingyu Guan, Anand Padmanabha Iyer, and Taesoo Kim. Dynagraph: Dynamic graph neural networks at scale. In Proceedings of the 5th ACM SIGMOD Joint International Workshop on Graph Data Management Experiences & Systems (GRADES) and Network Data Analytics (NDA), GRADES-NDA ’22, Ne...

    2022

  8. [16]

    Attention based spatial-temporal graph convolutional networks for traffic flow forecast- ing

    Shengnan Guo, Youfang Lin, Ning Feng, Chao Song, and Huaiyu Wan. Attention based spatial-temporal graph convolutional networks for traffic flow forecast- ing. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pages 922–929, 2019

    2019

  9. [17]

    In- ductive representation learning on large graphs

    Will Hamilton, Zhitao Ying, and Jure Leskovec. In- ductive representation learning on large graphs. In I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fer- gus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30, pages 102...

    2017

  10. [18]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Representation Learning on Graphs: Methods and Ap- plications. IEEE Data Engineering Bulletin , page arXiv:1709.05584, September 2017

    2017 arXiv

  11. [19]

    Long Short- Term Memory

    Sepp Hochreiter and Jürgen Schmidhuber. Long Short- Term Memory. Neural Computation, 9(8):1735–1780, 11 1997

    1997

  12. [21]

    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. CoRR, abs/2005.00687, 2020

    2005 arXiv

  13. [22]

    T-gcn: A sampling based streaming graph neural network system with hybrid ar- chitecture

    Chengying Huan, Shuaiwen Leon Song, Yongchao Liu, Heng Zhang, Hang Liu, Charles He, Kang Chen, Jin- lei Jiang, and Yongwei Wu. T-gcn: A sampling based streaming graph neural network system with hybrid ar- chitecture. In Proceedings of the International Confer- ence on Parallel...

  14. [23]

    Lsgcn: Long short-term traffic prediction with graph convolutional networks

    Rongzhou Huang, Chuyin Huang, Yubao Liu, Genan Dai, and Weiyang Kong. Lsgcn: Long short-term traffic prediction with graph convolutional networks. In IJCAI, volume 7, pages 2355–2361, 2020

    2020

  15. [24]

    ASAP: Fast, approximate graph pattern mining at scale

    Anand Padmanabha Iyer, Zaoxing Liu, Xin Jin, Shiv- aram Venkataraman, Vladimir Braverman, and Ion Sto- ica. ASAP: Fast, approximate graph pattern mining at scale. In 13th USENIX Symposium on Operating Systems Design and Implementation (OSDI 18), pages 745–761, Carlsbad, CA, Oc...

    2018

  16. [25]

    Gonzalez, and Ion Stoica

    Anand Padmanabha Iyer, Qifan Pu, Kishan Patel, Joseph E. Gonzalez, and Ion Stoica. TEGRA: Effi- cient Ad-Hoc analytics on evolving graphs. In 18th USENIX Symposium on Networked Systems Design and Implementation (NSDI 21), pages 337–355. USENIX Association, April 2021

    2021

  17. [26]

    H. V . Jagadish, Johannes Gehrke, Alexandros Labrini- dis, Yannis Papakonstantinou, Jignesh M. Patel, Raghu Ramakrishnan, and Cyrus Shahabi. Big data and its technical challenges. Commun. ACM, 57(7):86–94, jul 2014

    2014

  18. [27]

    Improving the accuracy, scalability, and performance of graph neural networks with roc

    Zhihao Jia, Sina Lin, Mingyu Gao, Matei Zaharia, and Alex Aiken. Improving the accuracy, scalability, and performance of graph neural networks with roc. In I. Dhillon, D. Papailiopoulos, and V . Sze, editors,Pro- ceedings of Machine Learning and Systems, volume 2, pages 187–19...

    2020

  19. [29]

    Ex- amining COVID-19 forecasting using spatio-temporal graph neural networks

    Amol Kapoor, Xue Ben, Luyang Liu, Bryan Perozzi, Matt Barnes, Martin Blais, and Shawn O’Banion. Ex- amining COVID-19 forecasting using spatio-temporal graph neural networks. CoRR, abs/2007.03113, 2020

    2007 arXiv

  20. [30]

    A fast and high qual- ity multilevel scheme for partitioning irregular graphs

    George Karypis and Vipin Kumar. A fast and high qual- ity multilevel scheme for partitioning irregular graphs. SIAM J. Sci. Comput., 20(1):359–392, December 1998

    1998

  21. [31]

    A fast and high qual- ity multilevel scheme for partitioning irregular graphs

    George Karypis and Vipin Kumar. A fast and high qual- ity multilevel scheme for partitioning irregular graphs. SIAM Journal on scientific Computing, 20(1):359–392, 1998

    1998

  22. [32]

    Representation learning for dynamic graphs: A survey, 2020

    Seyed Mehran Kazemi, Rishab Goel, Kshitij Jain, Ivan Kobyzev, Akshay Sethi, Peter Forsyth, and Pascal Poupart. Representation learning for dynamic graphs: A survey, 2020

    2020

  23. [33]

    Zipg: A memory-efficient graph store for interactive queries

    Anurag Khandelwal, Zongheng Yang, Evan Ye, Rachit Agarwal, and Ion Stoica. Zipg: A memory-efficient graph store for interactive queries. In Proceedings of the 2017 ACM International Conference on Management of Data, pages 1149–1164, 2017

    2017

  24. [34]

    Farzad Khorasani, Keval V ora, Rajiv Gupta, and Laxmi N. Bhuyan. Cusha: Vertex-centric graph pro- cessing on gpus. In Proceedings of the 23rd Interna- tional Symposium on High-Performance Parallel and Distributed Computing, HPDC ’14, pages 239–252, New York, NY , USA, 2014. As...

    2014

  25. [35]

    GRIP: A Graph Neural Network Accelerator Architec- ture

    Kevin Kiningham, Christopher Re, and Philip Levis. GRIP: A Graph Neural Network Accelerator Architec- ture. arXiv e-prints, page arXiv:2007.13828, July 2020

    2007 arXiv

  26. [36]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-Supervised Classification with Graph Convolutional Networks. In Proceedings of the 5th International Conference on Learning Representations, ICLR ’17, 2017

    2017

  27. [37]

    Howie Huang

    Pradeep Kumar and H. Howie Huang. GraphOne: A data store for real-time analytics on evolving graphs. In 17th USENIX Conference on File and Storage Technologies (FAST 19), pages 249–263, Boston, MA, February 2019. USENIX Association

    2019

  28. [38]

    Professor forcing: A new algorithm for training recurrent networks

    Alex M Lamb, Anirudh Goyal ALIAS PARTH GOY AL, Ying Zhang, Saizheng Zhang, Aaron C Courville, and Yoshua Bengio. Professor forcing: A new algorithm for training recurrent networks. Advances in neural information processing systems, 29, 2016

    2016

  29. [39]

    Dynamic graph convolutional recurrent network for traffic prediction: Benchmark and solution

    Fuxian Li, Jie Feng, Huan Yan, Guangyin Jin, Fan Yang, Funing Sun, Depeng Jin, and Yong Li. Dynamic graph convolutional recurrent network for traffic prediction: Benchmark and solution. ACM Trans. Knowl. Discov. Data, 17(1), feb 2023

    2023

  30. [40]

    Cache-based gnn system for dynamic graphs

    Haoyang Li and Lei Chen. Cache-based gnn system for dynamic graphs. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, CIKM ’21, page 937–946, New York, NY , USA, 2021. Association for Computing Machinery

    2021

  31. [41]

    Dif- fusion convolutional recurrent neural network: Data- driven traffic forecasting

    Yaguang Li, Rose Yu, Cyrus Shahabi, and Yan Liu. Dif- fusion convolutional recurrent neural network: Data- driven traffic forecasting. In International Conference on Learning Representations (ICLR ’18), 2018

    2018

  32. [42]

    Pagraph: Scaling gnn training on large graphs via computation-aware caching

    Zhiqi Lin, Cheng Li, Youshan Miao, Yunxin Liu, and Yinlong Xu. Pagraph: Scaling gnn training on large graphs via computation-aware caching. In Proceed- ings of the 11th ACM Symposium on Cloud Computing, SoCC ’20, pages 401–415, New York, NY , USA, 2020. Association for Computi...

    2020

  33. [43]

    Rensi, Wen Torng, and Russ B

    Yu-Chen Lo, Stefano E. Rensi, Wen Torng, and Russ B. Altman. Machine learning in chemoinformatics and drug discovery. Drug Discovery Today, 23(8):1538 – 1546, 2018

    2018

  34. [44]

    Kilmer, and Haim Avron

    Osman Asif Malik, Shashanka Ubaru, Lior Horesh, Misha E. Kilmer, and Haim Avron. Dynamic graph convolutional networks using the tensor m-product. In Proceedings of the 2021 SIAM International Confer- ence on Data Mining (SDM), pages 729–737. Society for Industrial and Applied ...

    2021

  35. [45]

    Dynamic graph convolutional networks

    Franco Manessi, Alessandro Rozza, and Mario Manzo. Dynamic graph convolutional networks. Pattern Recog- nition, 97:107000, Jan 2020

    2020

  36. [46]

    Graphbolt: Dependency-driven synchronous processing of stream- ing graphs

    Mugilan Mariappan and Keval V ora. Graphbolt: Dependency-driven synchronous processing of stream- ing graphs. In Proceedings of the Fourteenth EuroSys Conference 2019, EuroSys ’19, pages 25:1–25:16, New York, NY , USA, 2019. ACM

    2019

  37. [47]

    Marius: Learning massive graph embeddings on a single ma- chine

    Jason Mohoney, Roger Waleffe, Henry Xu, Theodoros Rekatsinas, and Shivaram Venkataraman. Marius: Learning massive graph embeddings on a single ma- chine. In 15th USENIX Symposium on Operating Sys- tems Design and Implementation (OSDI 21), pages 533–

  38. [48]

    Pinner- sage: Multi-modal user embedding framework for rec- ommendations at pinterest

    Aditya Pal, Chantat Eksombatchai, Yitong Zhou, Bo Zhao, Charles Rosenberg, and Jure Leskovec. Pinner- sage: Multi-modal user embedding framework for rec- ommendations at pinterest. In Proceedings of the 26th 15 ACM SIGKDD International Conference on Knowledge Discovery & Data ...

    2020

  39. [49]

    Transfer graph neural networks for pandemic forecasting

    George Panagopoulos, Giannis Nikolentzos, and Michalis Vazirgiannis. Transfer graph neural networks for pandemic forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 4838–4845, 2021

    2021

  40. [50]

    Schardl, and Charles E

    Aldo Pareja, Giacomo Domeniconi, Jie Chen, Tengfei Ma, Toyotaro Suzumura, Hiroki Kanezashi, Tim Kaler, Tao B. Schardl, and Charles E. Leiserson. Evolvegcn: Evolving graph convolutional networks for dynamic graphs, 2019

    2019

  41. [51]

    Estimating node impor- tance in knowledge graphs using graph neural networks

    Namyong Park, Andrey Kan, Xin Luna Dong, Tong Zhao, and Christos Faloutsos. Estimating node impor- tance in knowledge graphs using graph neural networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’19, pages 596–606...

    2019

  42. [52]

    Community dis- covery in dynamic networks: A survey

    Giulio Rossetti and Rémy Cazabet. Community dis- covery in dynamic networks: A survey. ACM Comput. Surv., 51(2), feb 2018

    2018

  43. [53]

    PyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neu- ral Machine Learning Models

    Benedek Rozemberczki, Paul Scherer, Yixuan He, George Panagopoulos, Alexander Riedel, Maria Aste- fanoaei, Oliver Kiss, Ferenc Beres, , Guzman Lopez, Nicolas Collignon, and Rik Sarkar. PyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neu- ral Machine Learning ...

    2021

  44. [54]

    Dysat: Deep neural representation learn- ing on dynamic graphs via self-attention networks

    Aravind Sankar, Yanhong Wu, Liang Gou, Wei Zhang, and Hao Yang. Dysat: Deep neural representation learn- ing on dynamic graphs via self-attention networks. In Proceedings of the 13th international conference on web search and data mining, pages 519–527, 2020

    2020

  45. [55]

    Structured sequence modeling with graph convolutional recurrent networks, 2016

    Youngjoo Seo, Michaël Defferrard, Pierre Van- dergheynst, and Xavier Bresson. Structured sequence modeling with graph convolutional recurrent networks, 2016

    2016

  46. [56]

    Accelerating dynamic graph analytics on gpus

    Mo Sha, Yuchen Li, Bingsheng He, and Kian-Lee Tan. Accelerating dynamic graph analytics on gpus. Proc. VLDB Endow., 11(1):107–120, September 2017

    2017

  47. [57]

    Foundations and modelling of dynamic networks using dynamic graph neural networks: A survey

    Joakim Skarding, Bogdan Gabrys, and Katarzyna Mu- sial. Foundations and modelling of dynamic networks using dynamic graph neural networks: A survey. CoRR, abs/2005.07496, 2020

    2005 arXiv

  48. [58]

    Session-based social recommendation via dynamic graph attention networks

    Weiping Song, Zhiping Xiao, Yifan Wang, Laurent Char- lin, Ming Zhang, and Jian Tang. Session-based social recommendation via dynamic graph attention networks. In Proceedings of the Twelfth ACM international con- ference on web search and data mining, pages 555–563, 2019

    2019

  49. [59]

    Session-based social recommendation via dynamic graph attention networks

    Weiping Song, Zhiping Xiao, Yifan Wang, Laurent Char- lin, Ming Zhang, and Jian Tang. Session-based social recommendation via dynamic graph attention networks. In Proceedings of the Twelfth ACM International Con- ference on Web Search and Data Mining, WSDM ’19, page 555–563, N...

    2019

  50. [60]

    Stokes, Kevin Yang, Kyle Swanson, Wen- gong Jin, Andres Cubillos-Ruiz, Nina M

    Jonathan M. Stokes, Kevin Yang, Kyle Swanson, Wen- gong Jin, Andres Cubillos-Ruiz, Nina M. Donghia, Craig R. MacNair, Shawn French, Lindsey A. Car- frae, Zohar Bloom-Ackermann, Victoria M. Tran, Anush Chiappino-Pepe, Ahmed H. Badran, Ian W. Andrews, Emma J. Chory, George M. Ch...

    2020

  51. [61]

    Dorylus: Affordable, scalable, and accurate GNN train- ing with distributed CPU servers and serverless threads

    John Thorpe, Yifan Qiao, Jonathan Eyolfson, Shen Teng, Guanzhou Hu, Zhihao Jia, Jinliang Wei, Keval V ora, Ravi Netravali, Miryung Kim, and Guoqing Harry Xu. Dorylus: Affordable, scalable, and accurate GNN train- ing with distributed CPU servers and serverless threads. In 15th...

    2021

  52. [62]

    Graph attention networks

    Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. In International Conference on Learning Representations, 2018

    2018

  53. [63]

    Pipad: Pipelined and parallel dynamic gnn training on gpus

    Chunyang Wang, Desen Sun, and Yuebin Bai. Pipad: Pipelined and parallel dynamic gnn training on gpus. In Proceedings of the 28th ACM SIGPLAN Annual Sympo- sium on Principles and Practice of Parallel Program- ming, PPoPP ’23, page 405–418, New York, NY , USA,

  54. [64]

    Flex- graph: A flexible and efficient distributed framework for gnn training

    Lei Wang, Qiang Yin, Chao Tian, Jianbang Yang, Rong Chen, Wenyuan Yu, Zihang Yao, and Jingren Zhou. Flex- graph: A flexible and efficient distributed framework for gnn training. In Proceedings of the Sixteenth Euro- pean Conference on Computer Systems, EuroSys ’21, page 67–82,...

    2021

  55. [65]

    Deep Graph Library: A Graph- Centric, Highly-Performant Package for Graph Neu- ral Networks

    Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu, Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang 16 Li, and Zheng Zhang. Deep Graph Library: A Graph- Centric, Highly-Performant Package for Graph Neu- ral Networks. arXiv...

    1909 arXiv

  56. [66]

    Association for Computing Machinery

  57. [67]

    Williams and David Zipser

    Ronald J. Williams and David Zipser. A learning al- gorithm for continually running fully recurrent neural networks. Neural Computation, 1(2):270–280, 1989

    1989

  58. [68]

    Fast and Accu- rate Optimizer for Query Processing over Knowledge Graphs, page 503–517

    Jingqi Wu, Rong Chen, and Yubin Xia. Fast and Accu- rate Optimizer for Query Processing over Knowledge Graphs, page 503–517. Association for Computing Ma- chinery, New York, NY , USA, 2021

    2021

  59. [69]

    GNNAdvisor: An adaptive and efficient runtime system for GNN ac- celeration on GPUs

    Yuke Wang, Boyuan Feng, Gushu Li, Shuangchen Li, Lei Deng, Yuan Xie, and Yufei Ding. GNNAdvisor: An adaptive and efficient runtime system for GNN ac- celeration on GPUs. In 15th USENIX Symposium on Operating Systems Design and Implementation (OSDI 21), pages 515–531. USENIX As...

    2021

  60. [70]

    Gnnlab: a factored system for sample-based gnn training over gpus

    Jianbang Yang, Dahai Tang, Xiaoniu Song, Lei Wang, Qiang Yin, Rong Chen, Wenyuan Yu, and Jingren Zhou. Gnnlab: a factored system for sample-based gnn training over gpus. In Proceedings of the Seventeenth European Conference on Computer Systems, EuroSys ’22, page 417–434, New Y...

    2022

  61. [71]

    Hamilton, and Jure Leskovec

    Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombat- chai, William L. Hamilton, and Jure Leskovec. Graph convolutional neural networks for web-scale recom- mender systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Dis- covery & Data Mining, KDD ...

    2018

  62. [72]

    How powerful are graph neural networks? CoRR, abs/1810.00826, 2018

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? CoRR, abs/1810.00826, 2018

    2018 arXiv

  63. [73]

    Agl: A scalable system for industrial-purpose graph machine learning

    Dalong Zhang, Xin Huang, Ziqi Liu, Jun Zhou, Zhiyang Hu, Xianzheng Song, Zhibang Ge, Lin Wang, Zhiqiang Zhang, and Yuan Qi. Agl: A scalable system for industrial-purpose graph machine learning. Proc. VLDB Endow., 13(12):3125–3137, aug 2020

    2020

  64. [74]

    Under- standing GNN computational graph: A coordinated com- putation, io, and memory perspective

    Hengrui Zhang, Zhongming Yu, Guohao Dai, Guyue Huang, Yufei Ding, Yuan Xie, and Yu Wang. Under- standing GNN computational graph: A coordinated com- putation, io, and memory perspective. In Diana Mar- culescu, Yuejie Chi, and Carole-Jean Wu, editors,Pro- ceedings of Machine Le...

    2022

  65. [75]

    Spatio- temporal graph convolutional neural network: A deep learning framework for traffic forecasting

    Bing Yu, Haoteng Yin, and Zhanxing Zhu. Spatio- temporal graph convolutional neural network: A deep learning framework for traffic forecasting. CoRR, abs/1709.04875, 2017

    2017 arXiv

  66. [76]

    Dynamic graph neural networks for sequential recommendation

    Mengqi Zhang, Shu Wu, Xueli Yu, Qiang Liu, and Liang Wang. Dynamic graph neural networks for sequential recommendation. IEEE Transactions on Knowledge and Data Engineering, 35(5):4741–4753, 2022

    2022

  67. [77]

    Exploring the hidden dimension in graph processing

    Mingxing Zhang, Yongwei Wu, Kang Chen, Xuehai Qian, Xue Li, and Weimin Zheng. Exploring the hidden dimension in graph processing. In OSDI, volume 16, pages 285–300, 2016

    2016

  68. [78]

    Gaan: Gated attention net- works for learning on large and spatiotemporal graphs

    Jiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, and Dit-Yan Yeung. Gaan: Gated attention net- works for learning on large and spatiotemporal graphs. arXiv preprint arXiv:1803.07294, 2018

    2018 arXiv

  69. [79]

    T-gcn: A tempo- ral graph convolutional network for traffic prediction

    Ling Zhao, Yujiao Song, Chao Zhang, Yu Liu, Pu Wang, Tao Lin, Min Deng, and Haifeng Li. T-gcn: A tempo- ral graph convolutional network for traffic prediction. IEEE Transactions on Intelligent Transportation Sys- tems, 21(9):3848–3858, Sep 2020

    2020

  70. [80]

    Tgl: A general framework for temporal gnn training on billion-scale graphs

    Hongkuan Zhou, Da Zheng, Israt Nisa, Vasileios Ioanni- dis, Xiang Song, and George Karypis. Tgl: A general framework for temporal gnn training on billion-scale graphs. Proc. VLDB Endow. , 15(8):1572–1580, apr 2022. 17

    2022

  71. [81]

    Egraph: Efficient concurrent gpu-based dynamic graph processing

    Yu Zhang, Yuxuan Liang, Jin Zhao, Fubing Mao, Lin Gu, Xiaofei Liao, Hai Jin, Haikun Liu, Song Guo, Yangqing Zeng, Hang Hu, Chen Li, Ji Zhang, and Biao Wang. Egraph: Efficient concurrent gpu-based dynamic graph processing. IEEE Transactions on Knowledge and Data Engineering, 35...

    2023

  72. [549]

    USENIX Association, July 2021

    2021

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.