Pith. sign in

REVIEW 5 major objections 7 minor 55 references

GraphEdge: Dynamic Graph Partition and Task Scheduling for GNNs Computing in Edge Network

T0 review · 5 major / 7 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read GraphEdge claims that cutting the user-association graph into weakly linked subgraphs and offloading each subgraph to a single edge server minimizes system cost and cross-server communication for GNN inference in dynamic edge networks.

desk verdict Plausible GNN edge-offloading architecture with a genuinely new partition-for-aggregation idea, but a factor-N inflation in the cost model undermines the central quantitative claim. read the letter →

arxiv 2504.15905 v1 pith:E3ARQX2Y submitted 2025-04-22 cs.LG cs.AI

Wenjing Xiao , Chenglong Shi , Miaojiang Chen , Zhiquan Liu , Min Chen , H. Herbert Song This is my paper · ORCID
classification cs.LGcs.AI
keywords edgecomputinggraphneuralnetworkstaskoffloadingpartitioningdeepreinforcementlearningmulti-agentdynamictopologycross-servercommunication
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

This paper tries to establish that the cost of serving graph neural network tasks on distributed edge servers can be minimized by treating the user graph itself as the object to optimize: first cut the graph into strongly associated subgraphs, then offload each subgraph as a single unit. The proposed GraphEdge architecture combines a breadth-first hierarchical cut (HiCut), which places cuts where associations between layers are weakest, with a multi-agent deep reinforcement learning offloader (DRLGO) trained to keep each subgraph on one server. If the claim is right, correlated IoT tasks such as traffic prediction and social recommendation can be served at the edge without paying the full message-passing communication cost between servers, and the system can re-plan as users join, leave, or move. The paper's simulations on citation graphs show lower system cost and lower cross-server communication than PPO, greedy, and random offloading baselines under dynamic user counts, associations, and positions.

What carries the argument

The load-bearing machinery is the pair HiCut plus DRLGO. HiCut is a hierarchical traversal graph cut: it runs a breadth-first search from an unchosen vertex, counts the edges between consecutive layers, and records a candidate cut whenever the edge count falls; when the count later rises, it closes a subgraph at the recorded layer, ensuring that vertices inside a subgraph are strongly associated and vertices across subgraphs are weakly associated. This directly targets GNN message passing, since aggregation requires a server to fetch neighbor features across cut edges. DRLGO is a multi-agent deep reinforcement learning offloader built on the MADDPG scheme (multi-agent deep deterministic policy gradient): each edge server is an agent that acts on local observations, a central critic trains them together, and the reward is the negative system cost plus a penalty proportional to the number of servers a subgraph is spread across. The dynamic graph model, a mask array for joining and leaving users plus position attributes for mobility, feeds both stages at each time step. Together they convert the integer offloading problem into graph partitioning followed by learned subgraph-to-server assignment.

What would settle it

Settle it by taking the objective apart: in Eq. (12) the inter-server transfer delay $T^{\mathrm{tran}}_{k,l}$ appears inside the sum over all $N$ users, so the same transfer is counted once per user rather than once per server pair. Rewriting the objective with the transfer summed over server pairs and retraining DRLGO, or measuring actual transfer latency on a two-server testbed under the same offloading decisions, would show whether the reported cost advantage is an artifact of the counting. If the corrected or measured costs put DRLGO on par with the baselines, the central claim collapses.

Watch

Extended reading notes

Core claim

The central claim is that GNN inference cost in an edge system splits into two manageable subproblems instead of one intractable integer program. The first subproblem is graph layout: because a GNN aggregates a user's features with its neighbors' features, any edge of the user graph that crosses server boundaries forces a data transfer. HiCut traverses the graph layer by layer and cuts where the number of edges between successive BFS layers dips, producing subgraphs whose inter-subgraph edges are few. The second subproblem is offloading: given those subgraphs, DRLGO, a multi-agent actor-critic policy with centralized training and distributed execution, assigns each subgraph to an edge server, with a reward that penalizes scattering a subgraph across servers. The paper argues that this two-step procedure minimizes the system cost $T_{\mathrm{all}} + I_{\mathrm{all}}$ and adapts to dynamic user states through a mask module and position attributes in the graph model; the experiments are offered as evidence that the architecture outperforms PPO-based, greedy, and random offloading in dynamic scenarios.

Load-bearing premise

The load-bearing assumption is that the paper's equations for time and energy (Eqs. 7-13) accurately describe what a real edge server would spend, since DRLGO is trained to minimize those equations and the experiments evaluate those same equations; if the model over- or under-counts transfer cost, the claimed savings are not real.

Editorial extensions

If this is right

  • A service operator can run GNN inference over geographically distributed servers while exchanging data only across subgraph boundaries; the amount of inter-server traffic is set by HiCut's cut quality, not by the raw number of user associations.
  • When users join, leave, or move, the mask and position attributes update the graph and the same trained DRLGO can re-plan the offloading, so the architecture does not need to retrain from scratch at every topology change.
  • The reported complexity advantage means the graph-cut stage remains usable as graphs grow: $O(N^2 + NE)$ versus the compared min-cut method's $O(V^2E)$, with the gap widening on dense graphs.
  • Because DRLGO's reward couples system cost with subgraph cohesion, the learned policy should transfer across GNN models that share the aggregation step, which the experiments support by testing GCN, GAT, GraphSAGE, and SGC.

Reading between the lines

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

  • Beyond the paper's stated claims, the cost objective in Eq. (12) sums the inter-server transfer time $T^{\mathrm{tran}}_{k,l}$ inside the loop over all $N$ users, so the same cross-server transfer appears to be counted once per user rather than once per server pair; a corrected objective could change both the training signal and the reported cost comparisons.
  • A natural test the paper does not run is to replace the citation graphs with a temporal network of real mobility or social interactions, where edges appear and disappear at fine time scales; the mask and position model would then face edge churn rather than resampled static graphs, and it is an open question whether HiCut's edge-count heuristic still finds good cuts.
  • The claim that cutting by BFS-layer edge-count dips is near-optimal is heuristic; comparing HiCut against exact or balanced graph-partition objectives on the same datasets would tell whether the communication savings come from the cut heuristic itself or from the subgraph-cohesive offloading reward.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. This paper proposes GraphEdge, an architecture for GNN inference in edge networks. It models user data associations as a dynamic graph, partitions the graph into weakly coupled subgraphs using a BFS-based hierarchical cut (HiCut), and then uses a MADDPG-based deep reinforcement learning algorithm (DRLGO) to offload whole subgraphs to edge servers. The objective is to minimize total time and energy, including upload, inter-server communication, and GNN computation. Experiments on CiteSeer, Cora, and PubMed against PPO, greedy, and random offloading baselines report lower system cost and lower cross-server cost, with additional results on convergence and an ablation. The paper claims that DRLGO 'can minimize system cost' even in dynamic scenarios and that HiCut 'can effectively reduce cross-server communication cost.'

Significance. The manuscript addresses a relevant and timely problem: serving GNN inference over distributed edge servers with dynamic user topology. The proposed architecture is cleanly decomposed into a dynamic graph model, a graph partition algorithm (HiCut), and a multi-agent reinforcement learning offloading algorithm (DRLGO), and the simulation study is self-contained in that all methods are evaluated on the same cost equations. If the cost model were correct, the results showing DRLGO below PPO, greedy, and random baselines, together with the ablation against DRL-only, would provide preliminary evidence that graph-aware partition plus subgraph-level offloading can reduce system cost and cross-server communication. However, the load-bearing cost model issues identified in the major comments—particularly the factor-N inflation of transfer costs in Eqs. (12)-(13), the undefined indices in Eqs. (7)-(8), and the unverified optimality of HiCut—mean that the quantitative results as presented do not support the central claims. The work is a reasonable starting point, but it needs substantial correction and re-evaluation.

major comments (5)
  1. [3.5, Eqs. (12)-(13)] In the definitions of T_all and I_all, the terms sum_{k,l≠k} T^tran_{k,l} and sum_{k,l≠k} I^com_{k,l} do not depend on the user index n, yet they are placed inside the outer sum over n=1..N. This counts the aggregate inter-server transfer time and energy N times, overweighting cross-server communication by a factor of N relative to upload and computation costs. Because DRLGO's reward is the negative of T_all + I_all and the experimental results are computed from these same formulas, both the training objective and the evaluation metric are distorted. The transfer terms should be moved outside the user sum, or a per-user share of the transfer cost should be defined. This is a load-bearing issue: the paper's central claim is that GraphEdge minimizes the true system cost, but the minimized quantity is not a faithful statement of that cost, and the overweighted transfer term can bias the learned policy toward collocation even when that increases true time and energy.
  2. [3.3, Eqs. (7)-(8)] The inter-server communication model is not well defined. In Eq. (7), x_{k→l}(t) is written as sum_{i=1}^N X_i · w_ik · e_ij, but the index j is free; a sum over j (or over edges (i,j)) with a condition on the destination server, such as w_jl = 1, is missing. In Eq. (8), I^com_{k,l}(t) contains e_ij(t), w_ik(t), w_jl(t), and X_i(t) with i and j unbound and no summation; as written it is a product of scalars for a single unspecified edge, not the total energy cost of communication between servers k and l. Without a precise definition, the objective in Eqs. (12)-(13) and the HiCut cost in Eq. (15) cannot be evaluated or reproduced. These equations need to be rewritten with explicit summations over the appropriate user pairs.
  3. [4.2-4.3, Eq. (15)] The paper asserts that HiCut finds a cut that 'ensures minimal inter-subgraph edges' and 'ensures that the correlation between subgraphs is minimized,' but no proof or counterexample-based argument is provided. The algorithm's decision rule compares edge counts between consecutive BFS layers and records local troughs; the output depends on the choice of the BFS starting vertex, and this local heuristic is not guaranteed to yield a global minimum cut. In addition, Eq. (15) places the minimization over G_sub inside the double sum over server pairs k and l, which would allow a different partition for each pair; the intended optimization is min_{G_sub} of the sum over pairs. The authors should either supply a correctness argument for the optimality claim or explicitly present HiCut as a heuristic, and they should correct Eq. (15).
  4. [6.3, Fig. 10] The experiment comparing different GNN models (GCN, GAT, GraphSAGE, SGC) does not support the claimed 'strong model adaptation capability' because the cost model in Eqs. (9)-(11) depends only on the number of layers, the feature dimensions S_k, the degrees, and the server speed f_k; the GNN architecture itself appears nowhere in the cost equations. The text states that all models have three layers and 64 neurons, so for a fixed dataset and offloading plan the computed system cost is identical across the four models. The results in Fig. 10 therefore cannot demonstrate any variation due to the GNN model, and the model-adaptation claim needs reformulation or a cost model that actually reflects the different operations of GCN, GAT, GraphSAGE, and SGC.
  5. [5.2, Eq. (25), and Table 2] The reward function in Eq. (25) includes the subgraph-grouping term R_sp with a weight constant ζ, but the value of ζ is not reported anywhere in Table 2 or the text. Since this term shapes the learned policy in addition to the system cost C = T_all + I_all, the training objective is not fully specified and the experiments are not reproducible without this value. The authors should report ζ and, ideally, include a sensitivity analysis over ζ.
minor comments (7)
  1. [Table 1] Table 1 lists σ_{k,l} twice, with slightly different descriptions; one entry should be removed.
  2. [Eq. (13)] In Eq. (13), the first term inside the user sum is written as I^up_{i,m}, while the outer sum uses index n; the indices should be consistent.
  3. [Eq. (15)] Eq. (15) has a typo in the upper limit of the second sum: 'm' should presumably be 'M'.
  4. [Section 4.2 and Algorithm 1] In Section 4.2, d_n is described as 'the number of edges in the current layer,' but in Algorithm 1 it is incremented for neighbors of current-layer vertices that are not in the subgraph; the definition should be stated precisely.
  5. [Section 6.2, Fig. 6] Section 6.2 compares HiCut to the min-cut method in [36] only in terms of runtime (Fig. 6); no comparison of cut quality (e.g., number of inter-subgraph edges or resulting communication cost) is reported, so the claim that HiCut reduces cross-server communication is not directly validated against the min-cut baseline.
  6. [Fig. 12] Fig. 12 appears garbled in the manuscript version I received (the figure body consists of '/uni0000...' tokens), so the ablation result comparing DRLGO and DRL-only cannot be verified from the text.
  7. [Section 6.3] The paper does not compare DRLGO to an optimal offloading solution on small instances or to a standard graph partitioner such as METIS; such comparisons would strengthen the claims that the offloading is 'optimal' and that HiCut has good cut quality.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the HiCut and DRLGO designs are self-contained; only minor non-load-bearing self-citations appear, and evaluation/training alignment is standard practice rather than a reduction.

full rationale

The paper's derivation chain is largely self-contained. HiCut is a BFS-based heuristic that cuts between BFS layers where the edge count decreases, and the cross-server communication cost is afterwards evaluated with the same cost model (Eqs. 7, 8, 12, 13) on all compared methods, so the comparison is fair across baselines. DRLGO is trained with a reward based on the system cost plus a subgraph-grouping penalty R_sp (Eqs. 23-25); reporting that the trained policy achieves low cost on the same simulator is a standard train/evaluation loop, not a fitted input renamed as a prediction, since DRLGO must still beat PPO, greedy, and random baselines on held-out dynamic scenarios. The abstract's claim that HiCut minimizes cross-server communication is a design consequence (cutting weakly associated subgraphs and offloading each subgraph to one server lowers inter-server edge traffic by construction), but the paper does not disguise this as an external first-principles theorem. Self-citations [2], [18], and [25] appear only in the introduction and related work as examples of IoT/EC research; none is load-bearing for the graph cut, the offloading algorithm, or the experimental claims. The undefined indices in Eq. (8) and the N-fold summation of T_tran and I_com in Eqs. (12)-(13) are correctness/validity concerns about the cost model, not circularity; they do not make the derivation equivalent to its inputs.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claims rest on a simplified simulation model (free-space channels, linear GNN energy) plus the unproven HiCut optimality criterion and an unreported reward weight zeta. The paper introduces no new physical entities, but its evaluation hinges on assumptions that are not validated against hardware.

free parameters (2)
  • zeta (subgraph grouping reward weight, Eq. 25)
    Hand-tuned weight controlling how strongly DRLGO is rewarded for placing a subgraph's users on the same server. Its value is absent from Table 2, so the trained policy cannot be reproduced.
  • GNN feature dimension cap for CiteSeer = 1500
    Dimensions greater than 1500 are truncated to 1500, which changes task data sizes for CiteSeer (3703 dims). This is a data preprocessing choice that affects the cost inputs.
assumptions (5)
  • domain assumption Wireless channel gain follows free-space path loss h = rho_0 d^{-2} (Eq. 3).
    Simplified channel model used to compute upload rates; ignores shadowing, fading, and interference beyond noise.
  • domain assumption The aggregate and update phases of GNN layers have per-unit energy costs mu, theta, phi, independent of server and data distribution (Eqs. 10-11).
    The GNN energy model is a linear abstraction of real inference energy.
  • ad hoc to paper A BFS layer whose edge count to the unassigned graph stops decreasing marks a boundary of weak association, so cutting there minimizes inter-subgraph edges.
    This heuristic optimality claim in Section 4.1-4.2 is unproven and is the basis for HiCut.
  • domain assumption Each AP is co-located with an edge server, so AP-server communication is free (Section 3.1).
    Used to ignore AP-server costs.
  • domain assumption All edge servers run identical, pre-trained GNN models; no model updates are transmitted (Section 3.1).
    The paper only considers inference, not training or model synchronization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GraphEdge: Dynamic Graph Partition and Task Scheduling for GNNs Computing in Edge Network." pith.science (2026). https://pith.science/paper/E3ARQX2Y

@misc{pith2026250415905,
  author       = {Pith},
  title        = {Pith review of: GraphEdge: Dynamic Graph Partition and Task Scheduling for GNNs Computing in Edge Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E3ARQX2Y}},
  note         = {Machine review of arXiv:2504.15905}
}
read the original abstract

With the exponential growth of Internet of Things (IoT) devices, edge computing (EC) is gradually playing an important role in providing cost-effective services. However, existing approaches struggle to perform well in graph-structured scenarios where user data is correlated, such as traffic flow prediction and social relationship recommender systems. In particular, graph neural network (GNN)-based approaches lead to expensive server communication cost. To address this problem, we propose GraphEdge, an efficient GNN-based EC architecture. It considers the EC system of GNN tasks, where there are associations between users and it needs to take into account the task data of its neighbors when processing the tasks of a user. Specifically, the architecture first perceives the user topology and represents their data associations as a graph layout at each time step. Then the graph layout is optimized by calling our proposed hierarchical traversal graph cut algorithm (HiCut), which cuts the graph layout into multiple weakly associated subgraphs based on the aggregation characteristics of GNN, and the communication cost between different subgraphs during GNN inference is minimized. Finally, based on the optimized graph layout, our proposed deep reinforcement learning (DRL) based graph offloading algorithm (DRLGO) is executed to obtain the optimal offloading strategy for the tasks of users, the offloading strategy is subgraph-based, it tries to offload user tasks in a subgraph to the same edge server as possible while minimizing the task processing time and energy consumption of the EC system. Experimental results show the good effectiveness and dynamic adaptation of our proposed architecture and it also performs well even in dynamic scenarios.

Figures

Figures reproduced from arXiv: 2504.15905 by the authors.

Figure 1
Figure 1. EC system for GNN tasks. EC networks to efficiently generate optimal graph offload￾ing decision based on optimized graph layout. Overall, our contributions are as follows: • We propose an efficient GNN-based edge computing architecture, which transmits graph data to different edge servers for GNN inference, achieving GNN pro￾cessing over multi-edge cloud collaboration. • We propose a hierarchical traversal graph cut… view at source ↗
Figure 2
Figure 2. The processing flow of the GNN-based EC system. 3.1. System overview We consider the GNN-based edge computing architec￾ture as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The Cut Process of HiCut. 4.2. The process of HiCut First of all, HiCut is based on the graph’s breadth￾first search (BFS) algorithm [44] and layer-by-layer traver￾sal graph layout . Although the graph’s depth-first search (DFS) algorithm [45] can also traverse the graph, it is limited by stack memory and difficult to assess the associations between the current vertex and vertices outside the recursion path, so we … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The training process of DRLGO. 𝜃 ′ 𝑚 and 𝜑 ′ 𝑚 use the soft update with the following formulas: 𝜃 ′ 𝑚 ← 𝜏𝜃𝑚 + (1 − 𝜏)𝜃 ′ 𝑚 , (31) 𝜑 ′ 𝑚 ← 𝜏𝜑𝑚 + (1 − 𝜏)𝜑 ′ 𝑚 , (32) where 𝜏 is the soft update factor. As shown in [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Vertices degree distribution for different datasets. Datasets in experiment. Three citation network datasets are considered in our experiments which are CiteSeer, Cora, and PubMed [53], [54]. Each dataset only has one citation network, and contains bag-of-words represe…
Figure 7
Figure 7. Figure 7: Dynamic performance of different methods on CiteSeer. 50 100 150 200 250 300 The Number of Users 0.0 0.2 0.4 0.6 0.8 Normalized System Cost DRLGO PTOM GM RM (a) The change of user’s number. 250 500 750 1000 1250 1500 1750 The Number of Users' Associations 0.4 0.6 0.8 1…
Figure 8
Figure 8. Figure 8: Dynamic performance of different methods on Cora. number of edge servers is 25. Experimental results in [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Dynamic performance of different methods on PubMed. (a) The system cost on CiteSeer. (b) The system cost on Cora. (c) The system cost on PubMed [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: The system cost of various methods on different datasets using different GNN models. a user task with its neighboring tasks to the same edge server or offload the user task to the nearest server. Next, we investigate the performance of each method on different dataset…
Figure 11
Figure 11. Figure 11: The comparison of cost convergence between DRLGO and PTOM during training. in both the number of users and their associations. All user tasks being offloaded and processed represent the comple￾tion of a training round. Since the training data is randomly sampled and d…
Figure 12
Figure 12. Figure 12: The comparison of system cost between DRLGO and DRL-only methods. To validate the effectiveness of HiCut in minimizing cross-server communication cost, we compared the system cost of DRLGO with and without HiCut, and the latter is called DRL-only, which is trained wit…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 52 canonical work pages

  1. [1]

    Secure and traceable multikey image retrieval in cloud-assisted internet of things.IEEE Internet of Things Journal, pages 1–1, 2024

    Tengfei Yang, Yuanyuan Li, Jiawei He, Zhiquan Liu, Fang Ren, Teng Wang, and Gaopan Hou. Secure and traceable multikey image retrieval in cloud-assisted internet of things.IEEE Internet of Things Journal, pages 1–1, 2024

    work page 2024

  2. [2]

    Alqahtani,andMinChen

    Wenjing Xiao, Xin Ling, Miaojiang Chen, Junbin Liang, Salman A. Alqahtani,andMinChen. Mvpoa:Alearning-basedvehicleproposal offloading for cloud-edge-vehicle networks.IEEE Internet of Things Journal, pages 1–1, 2024. Preprint submitted to Elsevier Page 15 of 17 Dynamic graph partition and task scheduling for GNNs computing in edge network

    work page 2024

  3. [3]

    Enabling balanced data deduplication in mobile edge computing.IEEE Trans- actionsonParallelandDistributedSystems ,34(5):1420–1431,2023

    RuikunLuo,HaiJin,QiangHe,SongWu,andXiaoyuXia. Enabling balanced data deduplication in mobile edge computing.IEEE Trans- actionsonParallelandDistributedSystems ,34(5):1420–1431,2023

    work page 2023

  4. [4]

    Ripple: Enabling decentralized data deduplication at the edge.IEEE Transactions on Parallel and Distributed Systems, 2024

    RuikunLuo,QiangHe,FeifeiChen,SongWu,HaiJin,andYunYang. Ripple: Enabling decentralized data deduplication at the edge.IEEE Transactions on Parallel and Distributed Systems, 2024

    work page 2024

  5. [5]

    Edge data deduplication under uncertainties: A robust optimization approach.IEEE Transactions on Parallel and Distributed Systems, 2024

    RuikunLuo,QiangHe,MengxiXu,FeifeiChen,SongWu,JingYang, Yuan Gao, and Hai Jin. Edge data deduplication under uncertainties: A robust optimization approach.IEEE Transactions on Parallel and Distributed Systems, 2024

    work page 2024

  6. [6]

    Influence maximization on social graphs: A survey

    Yuchen Li, Ju Fan, Yanhao Wang, and Kian-Lee Tan. Influence maximization on social graphs: A survey. IEEE Transactions on Knowledge and Data Engineering, 30(10):1852–1872, 2018

    work page 2018

  7. [7]

    IEEETransactions on Vehicular Technology, pages 1–16, 2023

    Zhiquan Liu, Lin Wan, Jingjing Guo, Feiran Huang, Xia Feng, Libo Wang,andJianfengMa.Ppru:Aprivacy-preservingreputationupdat- ingschemeforcloud-assistedvehicularnetworks. IEEETransactions on Vehicular Technology, pages 1–16, 2023

    work page 2023

  8. [8]

    Evit: Privacy-preserving image retrieval via encrypted vision transformer in cloud computing.IEEE Transactions on Circuits and Systems for Video Technology, 34(8):7467–7483, 2024

    QihuaFeng,PeiyaLi,ZhixunLu,ChaozhuoLi,ZefanWang,Zhiquan Liu,ChunhuiDuan,FeiranHuang,JianWeng,andPhilipS.Yu. Evit: Privacy-preserving image retrieval via encrypted vision transformer in cloud computing.IEEE Transactions on Circuits and Systems for Video Technology, 34(8):7467–7483, 2024

    work page 2024

Show all 55 references
  1. [9]

    Novel transformation deep learning model for electrocardio- gram classification and arrhythmia detection using edge computing

    YiboHan,PuHan,BoYuan,ZhengZhang,LuLiu,andJohnPanneer- selvam. Novel transformation deep learning model for electrocardio- gram classification and arrhythmia detection using edge computing. Journal of Grid Computing, 22(1):7, 2024

    2024

  2. [10]

    Real-time monitoring and analysisoftrackandfieldathletesbasedonedgecomputinganddeep reinforcement learning algorithm.Alexandria Engineering Journal, 114:136–146, 2025

    Xiaowei Tang, Bin Long, and Li Zhou. Real-time monitoring and analysisoftrackandfieldathletesbasedonedgecomputinganddeep reinforcement learning algorithm.Alexandria Engineering Journal, 114:136–146, 2025

    2025

  3. [11]

    Agricultural weed identification in images and videos by integrating optimized deep learning architecture on an edge computing technology

    Nitin Rai, Yu Zhang, Maria Villamil, Kirk Howatt, Michael Ostlie, and Xin Sun. Agricultural weed identification in images and videos by integrating optimized deep learning architecture on an edge computing technology. Computers and Electronics in Agriculture, 216:108442, 2024

    2024

  4. [12]

    Pflow: An end-to-end heterogeneous acceleration frame- work for cnn inference on fpgas.Journal of Systems Architecture, 150:103113, 2024

    Yi Wan, Xianzhong Xie, Lingjie Yi, Bo Jiang, Junfan Chen, and Yi Jiang. Pflow: An end-to-end heterogeneous acceleration frame- work for cnn inference on fpgas.Journal of Systems Architecture, 150:103113, 2024

    2024

  5. [13]

    Asurveyofvideo surveillance systems in smart city.Electronics, 12(17):3567, 2023

    YanjinlkhamMyagmar-OchirandWooseongKim. Asurveyofvideo surveillance systems in smart city.Electronics, 12(17):3567, 2023

    2023

  6. [14]

    mm-casgan: A cascaded adversarial neural framework for mmwave radar point cloud enhancement.Information Fusion, 108:102388, 2024

    Kareeb Hasan, Beng Oh, Nithurshan Nadarajah, and Mehmet Ra- sit Yuce. mm-casgan: A cascaded adversarial neural framework for mmwave radar point cloud enhancement.Information Fusion, 108:102388, 2024

    2024

  7. [15]

    Cool: a conjoint perspective on spatio-temporal graph neural network for traffic fore- casting

    WeiJu,YushengZhao,YifangQin,SiyuYi,JingyangYuan,Zhiping Xiao, Xiao Luo, Xiting Yan, and Ming Zhang. Cool: a conjoint perspective on spatio-temporal graph neural network for traffic fore- casting. Information Fusion, 107:102341, 2024

    2024

  8. [16]

    Eeoa: cost and energy efficient task scheduling in a cloud-fog framework.Sensors, 23(5):2445, 2023

    M Santhosh Kumar and Ganesh Reddy Karri. Eeoa: cost and energy efficient task scheduling in a cloud-fog framework.Sensors, 23(5):2445, 2023

    2023

  9. [17]

    A cost and energy efficient taskschedulingtechniquetooffloadmicroservicesbasedapplications in mobile cloud computing.IEEE Access, 10:46633–46651, 2022

    Abid Ali and Muhammad Munwar Iqbal. A cost and energy efficient taskschedulingtechniquetooffloadmicroservicesbasedapplications in mobile cloud computing.IEEE Access, 10:46633–46651, 2022

    2022

  10. [18]

    Adaptive compression offloading and resource allocationforedgevisioncomputing

    WenjingXiao,YixueHao,JunbinLiang,LongHu,SalmanAAlqah- tani, and Min Chen. Adaptive compression offloading and resource allocationforedgevisioncomputing. IEEETransactionsonCognitive Communications and Networking, 2024

    2024

  11. [19]

    Deep reinforcement learning for trajectory path planning and distributed inference in resource- constrained uav swarms

    MarwanAbdouDhuheir,EmnaBaccour,AimanErbad,SinanSabeeh Al-Obaidi, and Mounir Hamdi. Deep reinforcement learning for trajectory path planning and distributed inference in resource- constrained uav swarms. IEEE Internet of Things Journal , 10(9):8185–8201, 2022

    2022

  12. [20]

    Joint multi-domain resource allocation and trajectory optimization in uav-assistedmaritimeiotnetworks

    Li Ping Qian, Hongsen Zhang, Qian Wang, Yuan Wu, and Bin Lin. Joint multi-domain resource allocation and trajectory optimization in uav-assistedmaritimeiotnetworks. IEEEInternetofThingsJournal , 10(1):539–552, 2022

    2022

  13. [21]

    Digital twin-assisted urllc-enabled task offloading in mobile edge network via robust combinatorial optimization.IEEE Journal on Selected Areas in Communications, 2023

    Yixue Hao, Jiaxi Wang, Dongkun Huo, Nadra Guizani, Long Hu, and Min Chen. Digital twin-assisted urllc-enabled task offloading in mobile edge network via robust combinatorial optimization.IEEE Journal on Selected Areas in Communications, 2023

    2023

  14. [22]

    Joint sensing adaptation and model placement in 6g fabric computing.IEEE Journal on Selected Areas in Communications, 41(7):2013–2024, 2023

    Yixue Hao, Long Hu, and Min Chen. Joint sensing adaptation and model placement in 6g fabric computing.IEEE Journal on Selected Areas in Communications, 41(7):2013–2024, 2023

    2013

  15. [23]

    Collaborative computation offloading and resource allocation in multi-uav-assisted iotnetworks:Adeepreinforcementlearningapproach

    AbegazMohammedSeid,GordonOwusuBoateng,StephenAnokye, Thomas Kwantwi, Guolin Sun, and Guisong Liu. Collaborative computation offloading and resource allocation in multi-uav-assisted iotnetworks:Adeepreinforcementlearningapproach. IEEEInternet of Things Journal, 8(15):12203–12218, 2021

    2021

  16. [24]

    Jointuavtrajectoryplanning,dagtaskschedul- ing, andservice functiondeployment basedon drlin uav-empowered edge computing

    Xianglin Wei, Lingfeng Cai, Nan Wei, Peng Zou, Jin Zhang, and SureshSubramaniam. Jointuavtrajectoryplanning,dagtaskschedul- ing, andservice functiondeployment basedon drlin uav-empowered edge computing. IEEE Internet of Things Journal, 10(14):12826– 12838, 2023

    2023

  17. [25]

    Sgpl: An intelligent game-based secure collab- orative communication scheme for metaverse over 5g and beyond networks.IEEEJournalonSelectedAreasinCommunications ,2023

    Miaojiang Chen, Anfeng Liu, Neal N Xiong, Hongbing Song, and Victor CM Leung. Sgpl: An intelligent game-based secure collab- orative communication scheme for metaverse over 5g and beyond networks.IEEEJournalonSelectedAreasinCommunications ,2023

    2023

  18. [26]

    Model-assistedmulti-source fusion hypergraph convolutional neural networks for intelligent few- shot fault diagnosis to electro-hydrostatic actuator

    XiaoliZhao,XingjunZhu,JiahuiLiu,YuanhaoHu,TianyuGao,Liy- ongZhao,JianyongYao,andZhengLiu. Model-assistedmulti-source fusion hypergraph convolutional neural networks for intelligent few- shot fault diagnosis to electro-hydrostatic actuator. Information Fusion, 104:102186, 2024

    2024

  19. [27]

    Enhancement of traffic forecasting through graph neural network-based information fusion techniques.Information Fusion, 110:102466, 2024

    Shams Forruque Ahmed, Sweety Angela Kuldeep, Sabiha Jannat Rafa, Javeria Fazal, Mahfara Hoque, Gang Liu, and Amir H Gan- domi. Enhancement of traffic forecasting through graph neural network-based information fusion techniques.Information Fusion, 110:102466, 2024

    2024

  20. [28]

    Neural network operators: constructive interpola- tion of multivariate functions.Neural Networks, 67:28–36, 2015

    Danilo Costarelli. Neural network operators: constructive interpola- tion of multivariate functions.Neural Networks, 67:28–36, 2015

    2015

  21. [29]

    Spectral networks and locally connected networks on graphs.arXiv preprint arXiv:1312.6203, 2013

    Joan Bruna, Wojciech Zaremba, Arthur Szlam, and Yann LeCun. Spectral networks and locally connected networks on graphs.arXiv preprint arXiv:1312.6203, 2013

    2013 arXiv

  22. [30]

    Inductive repre- sentation learning on large graphs.Advances in Neural Information Processing Systems, 30, 2017

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive repre- sentation learning on large graphs.Advances in Neural Information Processing Systems, 30, 2017

    2017

  23. [31]

    Spatio- temporalfusiongraphconvolutionalnetworkfortrafficflowforecast- ing

    YingMa,HaijieLou,MingYan,FanghuiSun,andGuoqiLi. Spatio- temporalfusiongraphconvolutionalnetworkfortrafficflowforecast- ing. Information Fusion, 104:102196, 2024

    2024

  24. [32]

    Fograph:Enablingreal-timedeepgraphinferencewithfog computing

    Liekang Zeng, Peng Huang, Ke Luo, Xiaoxi Zhang, Zhi Zhou, and XuChen. Fograph:Enablingreal-timedeepgraphinferencewithfog computing. InProceedings of the ACM Web Conference 2022, pages 1774–1784, 2022

    2022

  25. [33]

    Fedgraphnn: A federated learning system and bench- mark for graph neural networks.arXiv preprint arXiv:2104.07145, 2021

    Chaoyang He, Keshav Balasubramanian, Emir Ceyani, Carl Yang, Han Xie, Lichao Sun, Lifang He, Liangwei Yang, Philip S Yu, Yu Rong, et al. Fedgraphnn: A federated learning system and bench- mark for graph neural networks.arXiv preprint arXiv:2104.07145, 2021

    2021 arXiv

  26. [34]

    Reducing com- munication in graph neural network training

    Alok Tripathy, Katherine Yelick, and Aydın Buluç. Reducing com- munication in graph neural network training. InSC20: International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1–14. IEEE, 2020

    2020

  27. [35]

    Distdgl: distributed graph neural network training for billion-scale graphs

    Da Zheng, Chao Ma, Minjie Wang, Jinjing Zhou, Qidong Su, Xiang Song, Quan Gan, Zheng Zhang, and George Karypis. Distdgl: distributed graph neural network training for billion-scale graphs. In 2020IEEE/ACM10thWorkshoponIrregularApplications:Architec- tures and Algorithms (IA3),...

    2020

  28. [36]

    Gnn at the edge: Cost-efficient graph neural network processing over distributed edge servers.IEEE Journal on Selected Areas in Communications, 41(3):720–739, 2022

    Liekang Zeng, Chongyu Yang, Peng Huang, Zhi Zhou, Shuai Yu, and Xu Chen. Gnn at the edge: Cost-efficient graph neural network processing over distributed edge servers.IEEE Journal on Selected Areas in Communications, 41(3):720–739, 2022

    2022

  29. [37]

    Intelligent offloading and resource allocation in heteroge- neous aerial access iot networks.IEEE Internet of Things Journal, 10(7):5704–5718, 2022

    Demeke Shumeye Lakew, Anh-Tien Tran, Nhu-Ngoc Dao, and Sun- grae Cho. Intelligent offloading and resource allocation in heteroge- neous aerial access iot networks.IEEE Internet of Things Journal, 10(7):5704–5718, 2022. Preprint submitted to Elsevier Page 16 of 17 Dynamic graph...

    2022

  30. [38]

    Multi-agent reinforcement learn- ing based resource management in mec-and uav-assisted vehicular networks

    Haixia Peng and Xuemin Shen. Multi-agent reinforcement learn- ing based resource management in mec-and uav-assisted vehicular networks. IEEE Journal on Selected Areas in Communications, 39(1):131–141, 2020

    2020

  31. [39]

    Graph-reinforcement- learning-basedtaskoffloadingformultiaccessedgecomputing

    Zhenchuan Sun, Yijun Mo, and Chen Yu. Graph-reinforcement- learning-basedtaskoffloadingformultiaccessedgecomputing. IEEE Internet of Things Journal, 10(4):3138–3150, 2021

    2021

  32. [40]

    Exploring graph neural networks for joint cruise control and task offloading in uav-enabled mobile edge computing

    Kai Li, Wei Ni, Xin Yuan, Alam Noor, and Abbas Jamalipour. Exploring graph neural networks for joint cruise control and task offloading in uav-enabled mobile edge computing. In2023 IEEE 97th Vehicular Technology Conference (VTC2023-Spring), pages 1–

  33. [41]

    Dynamic semantic compression for cnn inference in multi-access edge computing: A graph reinforcement learning-based autoencoder.IEEE Transactions on Wireless Communications, 2024

    Nan Li, Alexandros Iosifidis, and Qi Zhang. Dynamic semantic compression for cnn inference in multi-access edge computing: A graph reinforcement learning-based autoencoder.IEEE Transactions on Wireless Communications, 2024

    2024

  34. [42]

    Dependenttaskoffloadinginedgecomputingusinggnn and deep reinforcement learning.IEEE Internet of Things Journal, 2024

    Zequn Cao, Xiaoheng Deng, Sheng Yue, Ping Jiang, Ju Ren, and JinsongGui. Dependenttaskoffloadinginedgecomputingusinggnn and deep reinforcement learning.IEEE Internet of Things Journal, 2024

    2024

  35. [43]

    Zhi Wang, Tao Gong, Shu Hui Huang, and Bo Yi. Graph neural network with soft actor-critic and attention based large model for intelligent edge routing in consumer internet of things.IEEE Trans- actions on Consumer Electronics, 2025

    2025

  36. [44]

    Designing multithreaded algorithms for breadth-first search and st-connectivity on the cray mta-2

    David A Bader and Kamesh Madduri. Designing multithreaded algorithms for breadth-first search and st-connectivity on the cray mta-2. In 2006 International Conference on Parallel Processing (ICPP’06), pages 523–530. IEEE, 2006

    2006

  37. [45]

    Connectivity algorithm with depth first search (dfs) on simple graphs

    O Riansanti, M Ihsan, and D Suhaimi. Connectivity algorithm with depth first search (dfs) on simple graphs. In Journal of Physics: Conference Series, volume 948, page 012065. IOP Publishing, 2018

    2018

  38. [46]

    Multi-agent actor-critic for mixed cooperative- competitive environments.Advances in Neural Information Process- ing Systems, 30, 2017

    RyanLowe,YiIWu,AvivTamar,JeanHarb,OpenAIPieterAbbeel, and Igor Mordatch. Multi-agent actor-critic for mixed cooperative- competitive environments.Advances in Neural Information Process- ing Systems, 30, 2017

    2017

  39. [47]

    Multistageenergymanagement ofcoordinatedsmartbuildings:Amultiagentmarkovdecisionprocess approach

    Georgios Tsaousoglou, Nikolaos Efthymiopoulos, Prodromos Makris,andEmmanouelVarvarigos. Multistageenergymanagement ofcoordinatedsmartbuildings:Amultiagentmarkovdecisionprocess approach. IEEE Transactions on Smart Grid, 13(4):2788–2797, 2022

    2022

  40. [48]

    Data fusion and transfer learning empowered granular trust evaluation for internet of things.Information Fusion, 78:149–157, 2022

    Hui Lin, Sahil Garg, Jia Hu, Xiaoding Wang, Md Jalil Piran, and M Shamim Hossain. Data fusion and transfer learning empowered granular trust evaluation for internet of things.Information Fusion, 78:149–157, 2022

    2022

  41. [49]

    Fairness- based3-dmulti-uavtrajectoryoptimizationinmulti-uav-assistedmec system.IEEEInternetofThingsJournal ,10(13):11383–11395,2023

    Yejun He, Youhui Gan, Haixia Cui, and Mohsen Guizani. Fairness- based3-dmulti-uavtrajectoryoptimizationinmulti-uav-assistedmec system.IEEEInternetofThingsJournal ,10(13):11383–11395,2023

    2023

  42. [50]

    Chatmdg: A discourse parsing graph fusion based approach for multi-party dialogue generation

    JingyangLi,ShengliSong,YixinLi,HanxiaoZhang,andGuangneng Hu. Chatmdg: A discourse parsing graph fusion based approach for multi-party dialogue generation. Information Fusion, 110:102469, 2024

    2024

  43. [51]

    Simplifying graph convolutional networks

    Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simplifying graph convolutional networks. InInternationalConferenceonMachineLearning ,pages6861–6871. PMLR, 2019

    2019

  44. [52]

    Energy– latency tradeoff for computation offloading in uav-assisted multi- access edge computing system

    Kaiyuan Zhang, Xiaolin Gui, Dewang Ren, and Defu Li. Energy– latency tradeoff for computation offloading in uav-assisted multi- access edge computing system. IEEE Internet of Things Journal, 8(8):6709–6719, 2020

    2020

  45. [53]

    Revisiting semi-supervised learning with graph embeddings

    Zhilin Yang, William Cohen, and Ruslan Salakhudinov. Revisiting semi-supervised learning with graph embeddings. InInternational Conference on Machine Learning, pages 40–48. PMLR, 2016

    2016

  46. [54]

    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):93–93, 2008

    2008

  47. [55]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017. Preprint submitted to Elsevier Page 17 of 17

    2017 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.