Fully Distributed Online Training of Graph Neural Networks in Networked Systems

Ananthram Swami; Gunjan Verma; Kevin Chan; Rostyslav Olshevskyi; Santiago Segarra; Zhongyuan Zhao

REVIEW 3 major objections 5 minor 24 references

Fully Distributed Online Training of Graph Neural Networks in Networked Systems

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper shows that GNN training can run fully distributed with only L extra message-passing rounds per mini-batch, so networks can learn online without a central server.

desk verdict The local backprop derivation is real and the piggybacking idea is sensible, but the paper's headline communication-count claim currently rests on an off-by-two scheduling index that must be fixed before the paper is solid. read the letter →

arxiv 2412.06105 v1 pith:LE47QGI3 submitted 2024-12-08 cs.LG cs.DC

Rostyslav Olshevskyi , Zhongyuan Zhao , Kevin Chan , Gunjan Verma , Ananthram Swami , Santiago Segarra This is my paper · ORCID

classification cs.LGcs.DC

keywords graphneuralnetworksdistributedoptimizationonlinetraininglocalbackpropagationmessagepassingmini-batchwirelessdecentralizedlearning

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 claims that graph neural networks can be trained online in a fully distributed way, with no central server, by recasting training as a distributed optimization problem. For a mini-batch of $B$ samples and an $L$-layer graph convolutional network, the authors derive a local form of backpropagation and show that training costs only $L$ additional message-passing rounds beyond the $LB$ rounds already needed for inference, with messages about twice as large. They verify the approach numerically on supervised node regression, unsupervised power allocation, and reinforcement-learning link scheduling. If correct, this would let networked systems adapt their GNNs in deployment rather than retraining centrally in simulation.

What carries the argument

The key object is the recursive local backpropagation identity: $Z^{l-1}_{*i} = (\Theta^l_{i,0} + S_{ii}\Theta^l_{i,1})Q^l_{*i} + \sum_{j\in N_G(i)} S_{ji}\Theta^l_{j,1}Q^l_{*j}$, where the second term is the only one requiring messages from neighbors. This identity decomposes the global gradient into per-node terms and one exchange per layer, so each node can estimate its local gradient with $L-1$ rounds. On top of that, the paper uses three communication reductions: reusing forward-pass intermediates, running consensus once per mini-batch instead of per sample, and piggybacking the backward-pass messages of sample $b-1$ onto forward-pass messages of sample $b$.

What would settle it

Simulate the same training loop with a small probability of message loss or with edges changing mid-mini-batch, counting every attempted broadcast as a round; if the loss stops tracking centralized training or the required rounds to reach a target loss grow with the loss rate, the claimed $LB + L - 1$ overhead only holds in the idealized no-error setting.

Watch

Extended reading notes

Core claim

The paper's central claim is that the global gradient of a graph convolutional network can be computed from local quantities plus one neighbor broadcast per layer, which makes centralized SGD unnecessary. Concretely, for a node $i$, the backward pass needs only saved forward-pass activations and the products $Q^l_{*i}$; the only nonlocal term is the sum over neighbors of $S_{ji}\Theta^l_{j,1}Q^l_{*j}$, obtainable by one broadcast per node per layer. Combining this with consensus-based distributed optimizers, gradient aggregation over a mini-batch, and piggybacking of backward-pass messages onto the next forward pass yields $LB + L - 1$ rounds per mini-batch, exactly $L$ rounds more than inference, with message sizes roughly doubled. Numerical results on node regression, unrolled WMMSE power allocation, and distributed link scheduling show distributed training matching or closely approaching centralized training.

Load-bearing premise

The cost and convergence claims assume every message arrives correctly and the network connections do not change during one mini-batch, since the averaging weights are computed only once per mini-batch.

Editorial extensions

If this is right

Training a GNN no longer requires a central server; every node keeps a local copy of the shared parameters and updates it from neighbor messages.
The communication overhead of online training approaches that of inference: for a mini-batch of $B$ samples, only $L$ extra rounds are needed, so a network can continuously adapt its model in deployment.
Distributed optimizers such as D-SGD, D-Adam, and D-AMSGrad can be used with the local gradients; the experiments indicate D-AMSGrad tracks centralized training most closely when momentum is shared through consensus.
The derivation applies to GCNNs and the paper argues it extends to edge-featured GNNs and graph attention networks, as well as to unsupervised and reinforcement-learning pipelines built on GNNs.

Reading between the lines

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

The same message-piggybacking argument should extend to other local-message-passing GNN layers, such as graph attention networks, provided attention weights are reused consistently in backward passes; the paper states the principle carries over but does not test it.
The cost formula treats each broadcast as one round regardless of message size; in realistic wireless settings the doubled message size may split into extra transmission slots, so the true round-equivalent overhead could exceed $L$ for low-rate links.
Because topology is assumed static within a mini-batch, the method as presented would need rework to handle mobility or link failures online; the authors list these as future work, so a natural test is to rerun the link-scheduling experiment with a graph that changes mid-batch.
No theoretical convergence proof for the full piggybacked schedule is provided; the paper relies on existing distributed-optimization guarantees for the outer consensus loop, leaving open whether local backpropagation preserves those guarantees.

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.

Desk Editor's Note

read the letter

First, the stress-test concern is correct and important. In Section III-C, the piggybacking schedule uses \bar l = L-l-1, which for L=2 sends a layer-0 message and for general L sends backward messages in the wrong order. The backward message needed during the l-th forward round of the next sample is for layer L-l+1, not L-l-1. With that correction, the LB+L-1 count in Table I is achievable; as written, the schedule is not implementable. This is a revision-grade bug, not a wholesale invalidation, because the fix is a one-line index change.

What the paper does well: Section III-A derives a node-local backprop for GCNNs where each node's gradient depends on L-hop parameters, which is the genuine departure from standard distributed optimization. The algebra is internally consistent. The mini-batch rearrangement and piggybacking idea are sensible, and the communication table is coherent once the index is fixed. The three experiments are appropriate: synthetic node regression, UWMMSE power allocation, and actor-critic link scheduling, and they show distributed training tracking centralized within reasonable noise. Source code is linked. The citation pattern is fair; the authors use their own UWMMSE and scheduling frameworks because those are the pipelines being plugged into, not to inflate results.

Soft spots, in proportion: there is no convergence theory, and the paper explicitly defers it; for a method paper that is more a missing extension than a flaw. The experiments have no error bars or multiple seeds, so "very close to centralized" is a visual claim. The abstract says L added rounds while Table I says L-1; the latter is correct. The assumed reliable, static-topology communication is stated and acknowledged as future work, so the communication count is for an idealized setting.

This paper is for people working on distributed GNN training in wireless and networked systems. It deserves a serious referee, but with major revision. If the authors fix the index, align the abstract with Table I, and add seeds/error bars, I would be glad to cite it. I'd bring it to reading group to talk about how communication-cost claims can hide off-by-one errors.

Referee Report

3 major / 5 minor

Summary. The paper proposes a fully distributed, online training scheme for graph neural networks (GCNNs) in networked systems. It reformulates centralized GNN training as a distributed optimization problem using a local loss decomposition and a local form of backpropagation, then applies consensus-based distributed optimizers (D-SGD, D-Adam, D-AMSGrad). The main claimed contribution is a communication-efficient mini-batch implementation that piggybacks backward-pass messages of one sample onto forward-pass messages of the next, reducing the per-mini-batch message-passing rounds to LB + L - 1, compared with LB rounds for inference. The method is demonstrated in three settings: supervised node regression, unsupervised UWMMSE power allocation, and graph-based actor-critic link scheduling.

Significance. If the core communication-cost claim is correct, the paper addresses a real gap: GNN training over networks could be performed online in a fully distributed manner with overhead only linear in the number of GNN layers, rather than requiring centralized training or server-based subgraph partitioning. The local backpropagation derivation in Section III-A is clear and internally consistent, and the work is constructive and reproducible (source code and data are provided). The communication-cost table is a useful contribution. However, the headline claim is currently undermined by an indexing error in the piggybacking schedule, and the theoretical basis for the consensus-based optimizer is incomplete (the paper explicitly defers convergence proofs to future work). The experimental evidence also lacks multiple seeds or error bars.

major comments (3)

[Section III-C] The piggybacking schedule uses the backward-layer index \bar l = L-l-1 in the l-th round of the forward pass. This is incorrect. For L=2 and l=1 it gives \bar l=0, a non-existent layer; for general L it schedules backward messages for layers L-2, L-3, ..., 0 instead of the required order L, L-1, ..., 2. It also reverses the dependency order, because the backward message for a shallow layer cannot be produced before deeper-layer backward messages have been received from neighbors. Consequently, the claimed LB+L-1 rounds do not correspond to an implementable procedure as written. The correct mapping is \bar l = L-l+1 for l=1,...,L-1, which sends layer L, then L-1, and so on. The same correction must be applied to Table I, where the message size g_l + g_{L-l-1} should be g_l + g_{L-l}, since the piggybacked product \Theta^{\bar l}_{i1} Q^{\bar l}_{*i} has dimension g_{\bar l - 1} rather than g_{\bar l}.
[Abstract / Fig. 1 / Table I] The abstract states that training 'only adds L rounds of message passing to the LB rounds required by GNN inference,' and Fig. 1's caption says a mini-batch requires L(B+1) rounds; both expressions equal LB+L extra rounds. Table I and the text in Section III-C give LB+L-1, i.e., L-1 extra rounds. The derivation supports the LB+L-1 count. Please correct the abstract and figure caption, or explicitly state that the added rounds are at most L.
[Section III-B, Eqs. (8) and (17)] The quantity \nabla J_i(\theta) defined in (8c) is the i-th block of the global gradient, since it sums \partial \ell_j / \partial \theta_i over all j, not the gradient of the node-local objective J_i defined in (5). The consensus-based update (17) is therefore not the standard D-SGD/D-Adam/D-AMSGrad setting, and the convergence results cited from [14]–[16] do not directly apply to this gradient-block formulation. The paper defers convergence proofs to future work in Section V; please state explicitly what optimization problem (17) is intended to solve and under what assumptions the consensus-plus-block-gradient iteration is expected to converge, or soften the claim that the method is a direct application of those distributed optimizers.

minor comments (5)

[Section IV, all experiments] No error bars or multiple random seeds are reported in any of the three experiments. Single-run convergence curves are difficult to assess, especially in Fig. 2(a) where several methods perform similarly; please add multiple seeds and report variance, or state that the curves are representative runs.
[Section IV-C] The description says the actor GCNN is trained in a fully distributed manner while the critic GCNN is trained centrally. The abstract and contributions should qualify the reinforcement-learning experiment as a hybrid setup rather than fully distributed training of the whole pipeline.
[Section III-C, paragraph after Table I] The procedure for piggybacking the once-per-mini-batch transmission of d(i) and \theta_i(t) is vague. Please specify which message carries which part of \theta_i(t) and confirm that no additional message-passing round is introduced by this piggybacking.
[Eqs. (11) and (4)] The global loss in (4) includes the factor 1/|V|, but Z^L_{*i} in (11) is written without it. If this factor is absorbed into the learning rate, please say so explicitly to avoid a dimension mismatch in the gradient expressions.
[Notation, Eq. (8c)] The symbol \nabla J_i is misleading because (8c) is not the gradient of the local objective J_i. A distinct symbol, such as g_i or \nabla_{\theta_i} \ell, would prevent confusion and make the distributed-optimization formulation easier to follow.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the communication-round claims follow directly from the algorithm's message-passing structure and are benchmarked against centralized and external distributed baselines.

full rationale

The paper's central claim—that mini-batch GNN training can be done in LB + L − 1 message-passing rounds—is a direct consequence of the described procedure: L rounds of forward message passing per sample for B samples, L − 1 rounds for backward message passing of the final sample, and piggybacking of backward-pass messages on subsequent forward-pass messages (Section III-C). No parameter is fitted to a subset of data and then renamed a prediction; no uniqueness theorem is invoked to forbid alternatives; no defining quantity is defined in terms of the claimed result. The local backpropagation equations (11)–(14) are standard chain-rule identities derived in the text, and the consensus step uses known Metropolis-Hastings weights [21]. Self-citations [4] and [8] enter only as experimental testbeds for UWMMSE power allocation and link scheduling; they do not supply the theoretical communication-cost claim. The numerical comparisons are against centralized SGD and Adam baselines and against standard distributed optimizers D-SGD, D-Adam, and D-AMSGrad [14]–[16], so the method's convergence behavior is externally benchmarked. The reviewer-identified flaw in the piggybacking index ar{l} = L − l − 1 and the abstract/Table I discrepancy (L vs. L − 1 added rounds) are internal consistency and correctness concerns, not cases where an output equals an input by construction; they do not affect the circularity score.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central derivation is analytic: it decomposes backpropagation via the chain rule, so it introduces no fitted constants or invented quantities. The experimental sections use conventional hyperparameters (learning rates, batch sizes) that do not load-bear on the method's correctness. The main background assumptions are the domain assumptions listed above and the imported convergence results of the cited distributed optimizers.

assumptions (6)

domain assumption The graph G is connected and undirected, and the consensus matrix W from Metropolis-Hastings weights (16) is doubly stochastic.
Invoked for consensus updates (15) and (17); D-SGD/D-Adam/D-AMSGrad convergence results require doubly stochastic weights and connectivity.
domain assumption The network topology is static during the processing of a mini-batch of B samples.
Algorithm 1 computes W once per mini-batch (line 4) before the inner sample loop; the communication count in Table I assumes the same graph for all B samples.
domain assumption Each node executes synchronized message passing at every layer with no communication errors, and all messages are delivered.
Equations (10)-(14) presume lockstep broadcast and reception; Section V explicitly defers study of communication errors and mobility.
domain assumption The loss is a node-wise average (1c), so it can be decomposed as (4)-(5), and activations are differentiable so sigma'_l exists in (12).
Used in the loss decomposition and in the backpropagation equations (11)-(14).
standard math Standard chain rule and backpropagation calculus are valid for the GCNN architecture.
Used throughout Section III-A without independent proof.
domain assumption The cited D-SGD, D-Adam, and D-AMSGrad convergence guarantees (refs [14]-[16]) apply to the reformulated GNN objective with coupled local parameters.
The paper uses (17) to justify the distributed update but provides no convergence analysis of its own, explicitly deferring proofs to future work.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Fully Distributed Online Training of Graph Neural Networks in Networked Systems." pith.science (2026). https://pith.science/paper/LE47QGI3

@misc{pith2026241206105,
  author       = {Pith},
  title        = {Pith review of: Fully Distributed Online Training of Graph Neural Networks in Networked Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LE47QGI3}},
  note         = {Machine review of arXiv:2412.06105}
}

abstract

Graph neural networks (GNNs) are powerful tools for developing scalable, decentralized artificial intelligence in large-scale networked systems, such as wireless networks, power grids, and transportation networks. Currently, GNNs in networked systems mostly follow a paradigm of `centralized training, distributed execution', which limits their adaptability and slows down their development cycles. In this work, we fill this gap for the first time by developing a communication-efficient, fully distributed online training approach for GNNs applied to large networked systems. For a mini-batch with $B$ samples, our approach of training an $L$-layer GNN only adds $L$ rounds of message passing to the $LB$ rounds required by GNN inference, with doubled message sizes. Through numerical experiments in graph-based node regression, power allocation, and link scheduling in wireless networks, we demonstrate the effectiveness of our approach in training GNNs under supervised, unsupervised, and reinforcement learning paradigms.

Figures

Figures reproduced from arXiv: 2412.06105 by the authors.

**Figure 2.** The evolution of objective values over the course of training: (a) Node regression, where a marker is placed every 200 mini-batches. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 21 canonical work pages

[14]

Distributed stochastic gradient descent: Nonconvexity, nonsmoothness, and convergence to local minima,

B. Swenson, R. Murray, H. V . Poor, and S. Kar, “Distributed stochastic gradient descent: Nonconvexity, nonsmoothness, and convergence to local minima,” J. Mach. Learn. Res. , vol. 23, no. 328, pp. 1–62, 2022

work page 2022
[16]

On the convergence of decentralized adaptive gradient methods,

X. Chen, B. Karimi, W. Zhao, and P. Li, “On the convergence of decentralized adaptive gradient methods,” in Asian Conf. Mach. Learn. (ACML), vol. 189 of Proceedings of Machine Learning Research , pp. 217–232, PMLR, Dec 2023

work page 2023
[1]

A comprehensive survey on graph neural networks,

Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and S. Y . Philip, “A comprehensive survey on graph neural networks,” IEEE Trans. on Neural Networks and Learning Systems , 2020

work page 2020
[2]

Opportunities and challenges of graph neural networks in electrical engineering,

E. Chien, M. Li, A. Aportela, K. Ding, S. Jia, S. Maji, Z. Zhao, J. Duarte, V . Fung, C. Hao,et al., “Opportunities and challenges of graph neural networks in electrical engineering,” Nature Reviews Electrical Engineering, vol. 1, no. 8, pp. 529–546, 2024

work page 2024
[3]

Multi-agent deep reinforcement learning for dynamic power allocation in wireless networks,

Y . S. Nasir and D. Guo, “Multi-agent deep reinforcement learning for dynamic power allocation in wireless networks,” IEEE J. Sel. Areas Commun., vol. 37, no. 10, pp. 2239–2250, 2019

work page 2019
[4]

Unfolding WMMSE using graph neural networks for efficient power allocation,

A. Chowdhury, G. Verma, C. Rao, A. Swami, and S. Segarra, “Unfolding WMMSE using graph neural networks for efficient power allocation,” IEEE Trans. Wireless Commun. , vol. 20, no. 9, pp. 6004–6017, 2021

work page 2021
[5]

Link scheduling using graph neural networks,

Z. Zhao, G. Verma, C. Rao, A. Swami, and S. Segarra, “Link scheduling using graph neural networks,” IEEE Trans. Wireless Commun., vol. 22, no. 6, pp. 3997–4012, 2023

2023
[6]

Biased backpressure routing using link features and graph neural networks,

Z. Zhao, B. Radoji ˇci´c, G. Verma, A. Swami, and S. Segarra, “Biased backpressure routing using link features and graph neural networks,” IEEE Trans. Mach. Learn. Commun. Netw., vol. 2, pp. 1424–1439, 2024

work page 2024

Show all 24 references

[7]

Congestion-aware distributed task offloading in wireless multi-hop networks using graph neural networks,

Z. Zhao, J. Perazzone, G. Verma, and S. Segarra, “Congestion-aware distributed task offloading in wireless multi-hop networks using graph neural networks,” in IEEE Int. Conf. on Acoustics, Speech and Signal Process. (ICASSP), pp. 8951–8955, 2024

2024
[8]

Graph-based deterministic policy gradient for repetitive combinatorial optimization problems,

Z. Zhao, A. Swami, and S. Segarra, “Graph-based deterministic policy gradient for repetitive combinatorial optimization problems,” in Intl. Conf. Learn. Repres. (ICLR) , 2023

2023
[9]

Graph neural networks for wireless communications: From theory to practice,

Y . Shen, J. Zhang, S. H. Song, and K. B. Letaief, “Graph neural networks for wireless communications: From theory to practice,” IEEE Trans. Wireless Commun., vol. 22, no. 5, pp. 3554–3569, 2023

2023
[10]

GLANCE: Graph-based learnable digital twin for communication networks,

B. Li, G. Verma, T. Efimov, A. Kumar, and S. Segarra, “GLANCE: Graph-based learnable digital twin for communication networks,” arXiv preprint arXiv:2408.09040, 2024

2024 arXiv
[11]

A survey on distributed machine learning,

J. Verbraeken, M. Wolting, J. Katzy, J. Kloppenburg, T. Verbelen, and J. S. Rellermeyer, “A survey on distributed machine learning,” ACM Comput. Surv., vol. 53, Mar. 2020

2020
[12]

Communication-efficient learning of deep networks from decentralized data,

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y. Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Intl. Conf. Artif. Intel. Stat. (AISTATS), vol. 54, pp. 1273–1282, PMLR, Apr 2017

2017
[13]

Distributed subgradient methods for multi- agent optimization,

A. Nedic and A. Ozdaglar, “Distributed subgradient methods for multi- agent optimization,” IEEE Trans. Auto. Control , vol. 54, no. 1, pp. 48– 61, 2009

2009
[15]

DADAM: A consensus- based distributed adaptive gradient method for online optimization,

P. Nazari, D. A. Tarzanagh, and G. Michailidis, “DADAM: A consensus- based distributed adaptive gradient method for online optimization,” IEEE Trans. Signal Process. , vol. 70, pp. 6065–6079, 2022

2022
[17]

A comprehensive survey on distributed training of graph neural networks,

H. Lin, M. Yan, X. Ye, D. Fan, S. Pan, W. Chen, and Y . Xie, “A comprehensive survey on distributed training of graph neural networks,” Proceedings of the IEEE , vol. 111, no. 12, pp. 1572–1606, 2023

2023
[18]

Distributed graph neural network training: A survey,

Y . Shao, H. Li, X. Gu, H. Yin, Y . Li, X. Miao, W. Zhang, B. Cui, and L. Chen, “Distributed graph neural network training: A survey,” ACM Comput. Surv., vol. 56, Apr. 2024

2024
[19]

Convolutional neural network architectures for signals supported on graphs,

F. Gama, A. G. Marques, G. Leus, and A. Ribeiro, “Convolutional neural network architectures for signals supported on graphs,” IEEE Trans. Signal Process., vol. 67, no. 4, pp. 1034–1049, 2019

2019
[20]

Semi-supervised classification with graph convolutional networks,

T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” in Intl. Conf. Learn. Repres. (ICLR) , 2017

2017
[21]

Distributed average consensus with time- varying metropolis weights,

L. Xiao, S. Boyd, and S. Lall, “Distributed average consensus with time- varying metropolis weights,” Automatica, vol. 1, pp. 1–4, 2006

2006
[22]

Technical report 38.901: Study on channel model for frequencies from 0.5 to 100 GHz,

3rd Generation Partnership Project (3GPP), “Technical report 38.901: Study on channel model for frequencies from 0.5 to 100 GHz,” tech. rep., ETSI, 2020. Release 16

2020
[23]

On the complexity of scheduling in wireless networks,

C. Joo, G. Sharma, N. B. Shroff, and R. R. Mazumdar, “On the complexity of scheduling in wireless networks,” EURASIP J.Wireless Commun. and Netw. , vol. 2010, p. 418934, Sep 2010

2010
[24]

Local greedy approximation for scheduling in multihop wireless networks,

C. Joo and N. B. Shroff, “Local greedy approximation for scheduling in multihop wireless networks,”IEEE Trans. on Mobile Computing, vol. 11, no. 3, pp. 414–426, 2012

2012

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [14]

Distributed stochastic gradient descent: Nonconvexity, nonsmoothness, and convergence to local minima,

B. Swenson, R. Murray, H. V . Poor, and S. Kar, “Distributed stochastic gradient descent: Nonconvexity, nonsmoothness, and convergence to local minima,” J. Mach. Learn. Res. , vol. 23, no. 328, pp. 1–62, 2022

work page 2022

[2] [16]

On the convergence of decentralized adaptive gradient methods,

X. Chen, B. Karimi, W. Zhao, and P. Li, “On the convergence of decentralized adaptive gradient methods,” in Asian Conf. Mach. Learn. (ACML), vol. 189 of Proceedings of Machine Learning Research , pp. 217–232, PMLR, Dec 2023

work page 2023

[3] [1]

A comprehensive survey on graph neural networks,

Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and S. Y . Philip, “A comprehensive survey on graph neural networks,” IEEE Trans. on Neural Networks and Learning Systems , 2020

work page 2020

[4] [2]

Opportunities and challenges of graph neural networks in electrical engineering,

E. Chien, M. Li, A. Aportela, K. Ding, S. Jia, S. Maji, Z. Zhao, J. Duarte, V . Fung, C. Hao,et al., “Opportunities and challenges of graph neural networks in electrical engineering,” Nature Reviews Electrical Engineering, vol. 1, no. 8, pp. 529–546, 2024

work page 2024

[5] [3]

Multi-agent deep reinforcement learning for dynamic power allocation in wireless networks,

Y . S. Nasir and D. Guo, “Multi-agent deep reinforcement learning for dynamic power allocation in wireless networks,” IEEE J. Sel. Areas Commun., vol. 37, no. 10, pp. 2239–2250, 2019

work page 2019

[6] [4]

Unfolding WMMSE using graph neural networks for efficient power allocation,

A. Chowdhury, G. Verma, C. Rao, A. Swami, and S. Segarra, “Unfolding WMMSE using graph neural networks for efficient power allocation,” IEEE Trans. Wireless Commun. , vol. 20, no. 9, pp. 6004–6017, 2021

work page 2021

[7] [5]

Link scheduling using graph neural networks,

Z. Zhao, G. Verma, C. Rao, A. Swami, and S. Segarra, “Link scheduling using graph neural networks,” IEEE Trans. Wireless Commun., vol. 22, no. 6, pp. 3997–4012, 2023

2023

[8] [6]

Biased backpressure routing using link features and graph neural networks,

Z. Zhao, B. Radoji ˇci´c, G. Verma, A. Swami, and S. Segarra, “Biased backpressure routing using link features and graph neural networks,” IEEE Trans. Mach. Learn. Commun. Netw., vol. 2, pp. 1424–1439, 2024

work page 2024