Topology Generation of UAV Covert Communication Networks: A Graph Diffusion Approach with Incentive Mechanism

Fengshun Li; Qian Chen; Shaowen Qin; Wen Tian; Xiaohuan Li; Xin Tang; Yinqiu Liu; Youchun Gong

REVIEW 4 major objections 3 minor 15 references

Topology Generation of UAV Covert Communication Networks: A Graph Diffusion Approach with Incentive Mechanism

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A graph-diffusion policy plus a Stackelberg reward game can self-organize UAV networks into sparse, covert-friendly topologies.

desk verdict The GDPO and SG are not actually integrated: the topology reward has no covertness term and the SG update comes after the rewards are computed, so the paper's central claim does not follow from the algorithms. read the letter →

arxiv 2508.06746 v1 pith:SQURWTVD submitted 2025-08-08 cs.AI

Xin Tang , Qian Chen , Fengshun Li , Youchun Gong , Yinqiu Liu , Wen Tian , Shaowen Qin , Xiaohuan Li This is my paper · ORCID

classification cs.AI

keywords UAVnetworkscovertcommunicationtopologygenerationgraphdiffusionpolicyoptimizationStackelberggameincentivemechanismself-organizinggenerativeAI

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 a framework for self-organizing UAV networks that must keep communicating while staying hidden from eavesdroppers. It combines Graph Diffusion-based Policy Optimization (GDPO) with a Stackelberg Game (SG) incentive mechanism: the diffusion model generates sparse but well-connected network topologies, and the game pays UAVs for relay behavior that balances throughput, energy, and detection risk. The paper argues this integrated design lets the network adapt to moving nodes and changing ground-user demand while sustaining cooperation among self-interested UAVs. If correct, it gives a practical route to covert multi-UAV communication without centralized topology control.

What carries the argument

The load-bearing machinery is the GDPO denoising policy paired with the backward-induction Stackelberg solution. GDPO is graph diffusion policy optimization: a generative diffusion model that reconstructs a graph from noise, with an Eager Policy Gradient estimator (Eq. 14) that carries the reward of the final topology to every sampled denoising step, reducing gradient variance relative to REINFORCE. The SG models Alice as the leader setting a per-UAV reward $r_j$ and each UAV as a follower choosing transmit power $P_{txj}$; the unique equilibrium is found by backward induction, with Alice's utility $V_j$ concave when $Z^* \le 0$. The multi-objective reward (Eq. 13) ties graph generation to o

What would settle it

Run the paper's Table I scenario and compute the equilibrium reward $r^*$ from Eq. (11); if $r^* < \phi_j(\epsilon/g_j^2 + 1)$, the claimed unique Stackelberg equilibrium is not established. Separately, measure the average Willie detection probability $\bar{p}_j$ over fifty GDPO-generated topologies and compare it with DDPO/PPO topologies; if GDPO does not reduce $\bar{p}_j$ while keeping connectivity, the covertness claim is unsupported.

Watch

Extended reading notes

Core claim

The central claim is that topology generation and covert-behavior incentives can be solved as one loop rather than two separate problems. GDPO treats the UAV network as a graph and learns a denoising policy that reconstructs good topologies from noise, scoring candidate graphs by coverage, energy, connectivity, and redundant-coverage penalties. The SG mechanism then determines the reward Alice pays each UAV and the transmission power each UAV chooses; the paper proves a unique Stackelberg equilibrium exists under the concavity condition and uses that equilibrium to update the UAV state during topology training. The experiments compare GDPO with PPO and DDPO in a simulated urban low-altitude

Load-bearing premise

The uniqueness of the game equilibrium depends on the unproven condition $Z^* \le 0$, which requires Alice's equilibrium reward to satisfy $r^* \ge \phi_j(\epsilon/g_j^2 + 1)$; if the utility-maximizing reward falls below that threshold, Alice's payoff need not be concave and the unique-equilibrium proof does not go through.

Editorial extensions

If this is right

If the framework works as claimed, a UAV swarm can form and repair its own multi-hop topology without centralized control, adapting to changing node distributions and ground-user demand.
Sparsity becomes a security feature: removing redundant links and avoiding over-centralized hubs reduces the structural clues that eavesdroppers can use for traffic analysis and link inference.
The Stackelberg equilibrium gives Alice a predictable way to budget rewards: with a fixed maximum budget, there is an optimal number of relay UAVs beyond which utility declines.
The same game-theoretic reward can serve as the training signal for the diffusion policy, so topology generation and covert-incentive design improve in one learning loop.
The framework is designed to preserve full network connectivity even with dynamic mobility, since the connectivity reward explicitly penalizes any topology with more than one connected component.

Reading between the lines

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

Editorial inference: because the paper's Willie model detects only physical-layer transmissions, the topology-sparsity benefit is likely to matter most against traffic-analysis and link-inference adversaries; adding link-degree or centrality penalties to $r_{topo}$ would make that explicit.
Editorial inference: the SG plus generative-topology loop is not limited to UAVs; the same leader-reward/follower-power structure applies to any battery-limited multi-hop network, such as ground IoT or vehicle platoons, where nodes are self-interested.
Editorial inference: a natural testable extension is to include Willie's detection probability $p_j$ directly in the topology reward and then compare the covertness of GDPO-generated topologies against PPO/DDPO in a channel-aware simulator.

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.

Desk Editor's Note

read the letter

Dear colleague,

The paper combines Graph Diffusion Policy Optimization (GDPO, from the authors' NeurIPS 2024 paper) with a Stackelberg game (SG) for UAV covert network topology generation. The combination is new, and the writing is mostly clear. But the integration is only superficial, and the paper's central claim about covert communication doesn't hold up.

The main problem: the topology reward (Eq. 13) contains no covertness term—only coverage, energy, connectivity, and overlap. In Algorithm 2, the topology rewards are computed before the SG update (line 7), and the EPG gradient (Eq. 14) uses only those pre-update rewards. So the topology generator never sees Willie's detection probability or any covertness signal. Also, the SG's policy set is only (P_tx, r)—no edge or relay variables—so it cannot "guide relay behaviors and neighbor links" as the abstract claims. The load-bearing assertion that the framework "enhances covert communication" is not realized by the presented algorithms.

What is genuinely new is the specific application of GDPO to UAV topology generation with a multi-objective reward, and the idea of coupling this with a game-theoretic incentive mechanism. The system model is clear, and the utility functions are reasonable. The paper honestly cites its prior GDPO work rather than re-deriving it.

The soft spots, in order of severity. First, Theorem 1 is essentially asserted: uniqueness of the SE depends on the unproven condition Z* <= 0, and the proof simply says Alice has a unique optimal policy. Second, Willie's detection model shifts from energy detection to coherent detection "to simplify," with no justification. Third, the experiments show convergence curves and one utility plot, but no direct covertness metrics, no error bars, no code, and many free parameters are unspecified. The empirical support for "enhancement of covert communication" is missing.

This is a competent application paper, but the integration is broken. The fix would be to include a detection-probability term in the topology reward and to feed the SG-optimal transmit powers into the reward computation before the policy gradient update. If that were done, the paper could be useful. As it stands, it overclaims.

I'd still send it to peer review—a referee would spot these issues and the revision could fix them. But I wouldn't cite it in its current form.

Referee Report

4 major / 3 minor

Summary. The paper proposes a self-organizing UAV network framework that combines Graph Diffusion-based Policy Optimization (GDPO) with a Stackelberg Game (SG)-based incentive mechanism for covert communication. The SG models Alice as a leader and UAVs as followers, deriving an optimal transmit power and reward policy. GDPO is used to generate network topologies by optimizing a reward composed of coverage, energy, connectivity, and overlap terms. Experiments compare GDPO with PPO and DDPO in terms of topology reward and convergence, and show Alice's utility under different incentive schemes. The central claim is that the integrated framework generates sparse, well-connected topologies while guiding self-interested UAVs to choose relay behaviors and neighbor links that enhance covert communication.

Significance. The application of graph diffusion models to UAV topology generation is a plausible and timely idea, and the combination with an incentive mechanism could be relevant to covert UAV networks. If the framework truly integrated topology learning with covertness-aware incentives, it would represent a useful step. However, the manuscript as written does not deliver this integration: the topology reward contains no covertness term, and the SG operates on transmit power/reward only, not on link or relay actions. The theoretical basis is also incomplete. The paper does credit and build on prior GDPO work (Ref. [10]), but the claimed novelty depends on the integration, which is not demonstrated. Given these gaps, the current significance is limited to combining two existing components without a verifiable joint effect.

major comments (4)

[Section III, after Eq. (12), Theorem 1] The proof of Theorem 1 is conditional on the unproven assumption Z* ≤ 0, i.e., r ≥ φ_j(ε/g_j^2 + 1). The second derivative in Eq. (12) is strictly negative only under this inequality, and no argument is given that the equilibrium reward satisfies it. Without this condition, V_j may not be concave and the claimed unique Stackelberg equilibrium is not established. The proof is also a sketch: it asserts that Alice has a unique optimal policy given the UAVs' best responses, but provides no argument. A rigorous proof or an explicit verification step is needed.
[Section IV-A, Eq. (13), and Algorithm 2] The topology reward r_topo in Eq. (13) contains no covertness-related term; it depends only on coverage, energy, connectivity, and overlap. In Algorithm 2, r_k is computed from r_topo(S_k0) at line 5, before the state is updated using Algorithm 1 at line 7, and the EPG gradient (Eq. (14)) uses these pre-update rewards. Thus the SG transmit-power/reward solution from Algorithm 1 does not feed into the topology policy gradient, and Willie's detection probability appears nowhere in the topology learning objective. Consequently, the paper's central claim that the SG 'guides self-interested UAVs to choose relay behaviors and neighbor links that support ... covert communication' is not realized by the presented algorithms. The SG policy set contains only (P_tx_j, r_j), not edge or relay-choice variables, so it cannot directly guide neighbor-link selection.
[Section II-C, Eqs. (4)-(5)] The Willie detection model is internally inconsistent. The text states that Willie uses an energy detector, defines Z = ε − P_tx_j g_j^2, and then gives p_j = Q(Z/σ_noise) in Eq. (5). This is the form for coherent detection, not for energy detection under the Gaussian approximation described. The mean and variance of |Y|^2 under H1 also appear mis-specified (the variance should include the fluctuation of the signal term; the expression shown matches a noncentral chi-square approximation but is not derived). This inconsistency propagates into Alice's utility in Eq. (6) and the subsequent SG analysis, undermining the covertness model.
[Section V, Figs. 3-5] The experiments do not substantiate the claim of enhanced covert communication. Fig. 3 compares average topology reward, Fig. 4 shows a generated topology, and Fig. 5 compares Alice's utility under incentive schemes, but there is no metric such as detection probability, covert throughput, or signal-to-noise ratio at Willie. The baseline incentive schemes ('throughput-priority' and 'cost-priority') are not defined, no error bars or statistical significance are provided, and the number of seeds is not reported. Therefore the conclusion that the integrated framework improves covertness is not demonstrated by the presented results.

minor comments (3)

[Throughout] Notation is inconsistent, e.g., 'UA V' appears with spacing errors, and the symbol R is used both as the coverage radius and as the reward variable in Eq. (3). Please unify notation.
[Equation (11)] The derivative after substituting P*_txj appears to contain a sign issue: the last terms should be checked carefully. The expression currently reads '− ln(r_j/φ_j) − 1', but differentiating r_j ln(P*_txj) with P*_txj = r_j/φ_j − 1 yields a different combination. Please verify.
[References] Reference [5] is incomplete ('arXiv preprint arXiv:2408.11398' with no year or venue formatting) and several arXiv preprints are cited without publication status. Please complete all references.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the GDPO+SG construction is self-contained; the unproven Z*<=0 condition and the missing covertness term in the topology reward are correctness/integration gaps, not circular reductions.

full rationale

I walked the derivation chain. The SG section derives Uj concavity (Eqs. 8-9), the best response P* (Eq. 10), and Alice's optimality conditions (Eqs. 11-12). The uniqueness theorem is a backward-induction argument conditional on Vj concavity. The only gap there is that the inequality Z*<=0 is asserted, not proven; that is an omitted verification, not a definitional equivalence, so it is not circularity. The GDPO component is imported from the authors' prior NeurIPS work [10], including co-author Yinqiu Liu, but the paper states the EPG update explicitly (Eq. 14) and benchmarks against PPO and DDPO, so the central learning claim has independent empirical grounding rather than reducing to a self-citation. The topology reward (Eq. 13) has no Willie detection-probability term, and Algorithm 2 computes rewards before the SG state update; consequently the claimed covert-communication enhancement is not actually fed into the topology gradient. This is an effectiveness/coherence mismatch, not a case of a prediction being equivalent to its input by construction. No fitted parameter is renamed as a prediction, no known result is merely relabeled, and no uniqueness theorem is imported solely from the authors' prior work to forbid alternatives. I therefore find no significant circularity.

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

The paper introduces many hand-chosen constants in the reward and utility functions that are not measured or justified. It also relies on the prior GDPO method without verification and on a conditional concavity condition that is not proven. Together, these free parameters and assumptions mean the reported performance is highly tuned to the specific setup and may not generalize.

free parameters (4)

mu, psi, omega
Weights in Alice's utility (Eq. 6) are not specified in Table I or the text.
alpha, beta, gamma, delta
Weights in the topology reward (Eq. 13) are not listed or justified.
a, b, o, theta (vartheta), rho (varrho)
Constants in the energy reward (Section IV-A) are not provided.
phi, lambda
Environment constants in the LoS probability (Eq. 1) are not given.

assumptions (4)

domain assumption The Air-to-Ground channel model from [11] applies to the simulated urban scenario.
Section II-A adopts the model without validation in this setting.
domain assumption Willie's detection statistic |Y|^2 can be approximated as Gaussian, and his detector is coherent.
Section II-C assumes a Gaussian approximation via the central limit theorem and then switches to coherent detection without justification.
ad hoc to paper The condition Z* <= 0 (r >= phi_j(epsilon/g_j^2 + 1)) holds at the Stackelberg equilibrium.
Section III uses this inequality to assert concavity of Alice's utility, but it is not derived from the model.
domain assumption The Graph Diffusion Policy Optimization method from [10] is effective for UAV topology generation.
Section IV-B imports GDPO as a black box from prior work by the same research group.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Topology Generation of UAV Covert Communication Networks: A Graph Diffusion Approach with Incentive Mechanism." pith.science (2026). https://pith.science/paper/SQURWTVD

@misc{pith2026250806746,
  author       = {Pith},
  title        = {Pith review of: Topology Generation of UAV Covert Communication Networks: A Graph Diffusion Approach with Incentive Mechanism},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SQURWTVD}},
  note         = {Machine review of arXiv:2508.06746}
}

read the original abstract

With the growing demand for Uncrewed Aerial Vehicle (UAV) networks in sensitive applications, such as urban monitoring, emergency response, and secure sensing, ensuring reliable connectivity and covert communication has become increasingly vital. However, dynamic mobility and exposure risks pose significant challenges. To tackle these challenges, this paper proposes a self-organizing UAV network framework combining Graph Diffusion-based Policy Optimization (GDPO) with a Stackelberg Game (SG)-based incentive mechanism. The GDPO method uses generative AI to dynamically generate sparse but well-connected topologies, enabling flexible adaptation to changing node distributions and Ground User (GU) demands. Meanwhile, the Stackelberg Game (SG)-based incentive mechanism guides self-interested UAVs to choose relay behaviors and neighbor links that support cooperation and enhance covert communication. Extensive experiments are conducted to validate the effectiveness of the proposed framework in terms of model convergence, topology generation quality, and enhancement of covert communication performance.

Figures

Figures reproduced from arXiv: 2508.06746 by the authors.

**Figure 1.** illustrates a decentralized multi-UAV network designed to support high persistent connection and emergency networking of sensitive or covert data for Ground Users (GUs). The UAVs form a multi-hop communication topology to relay classified information from GUs to an aerial base station (such as an airship) while maintaining covert communication. The practical application scenarios include hydraulic and structural in… view at source ↗

**Figure 2.** UAV network topology optimization method based on GDPO. In output topologies, blue dashed lines represent redundant [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Topology generation [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 13 canonical work pages

[10]

Graph diffu- sion policy optimization,

Y . Liu, C. Du, T. Pang, C. Li, M. Lin, and W. Chen, “Graph diffu- sion policy optimization,” Advances in Neural Information Processing Systems, vol. 37, pp. 9585–9611, 2024

work page 2024
[1]

Digital-twin- assisted task assignment in multi-uav systems: A deep reinforcement learning approach,

X. Tang, X. Li, R. Yu, Y . Wu, J. Ye, F. Tang, and Q. Chen, “Digital-twin- assisted task assignment in multi-uav systems: A deep reinforcement learning approach,” IEEE Internet of Things Journal , vol. 10, no. 17, pp. 15 362–15 375, 2023

work page 2023
[2]

Uav-enabled communication strategy against detection in covert communication with asymmetric information,

W. Tian, J. Du, X. Ji, M. Du, G. Liu, and Z. Han, “Uav-enabled communication strategy against detection in covert communication with asymmetric information,” IEEE Transactions on Wireless Communica- tions, vol. 23, no. 7, pp. 6921–6937, 2023

work page 2023
[3]

Dynamic topology organization and maintenance algorithms for autonomous uav swarms,

A. Gaydamaka, A. Samuylov, D. Moltchanov, M. Ashraf, B. Tan, and Y . Koucheryavy, “Dynamic topology organization and maintenance algorithms for autonomous uav swarms,” IEEE Transactions on Mobile Computing, vol. 23, no. 5, pp. 4423–4439, 2023

work page 2023
[4]

Throughput maximiza- tion for covert uav relaying system,

H. Wang, H. Bai, F. Li, J. Chen, and L. Wang, “Throughput maximiza- tion for covert uav relaying system,” IEEE Transactions on V ehicular Technology, vol. 73, no. 3, pp. 4429–4434, 2023

work page 2023
[5]

Generative ai based secure wireless sensing for isac networks. arxiv 2024,

J. Wang, H. Du, Y . Liu, G. Sun, D. Niyato, S. Mao, D. Kim, and X. Shen, “Generative ai based secure wireless sensing for isac networks. arxiv 2024,” arXiv preprint arXiv:2408.11398

arXiv 2024
[6]

Covert communications: A comprehensive survey,

X. Chen, J. An, Z. Xiong, C. Xing, N. Zhao, F. R. Yu, and A. Nal- lanathan, “Covert communications: A comprehensive survey,” IEEE Communications Surveys & Tutorials , vol. 25, no. 2, pp. 1173–1198, 2023

work page 2023
[7]

Dnn task assignment in uav networks: A generative ai enhanced multi-agent reinforcement learning approach,

X. Tang, Q. Chen, W. Weng, B. Liao, J. Wang, X. Cao, and X. Li, “Dnn task assignment in uav networks: A generative ai enhanced multi-agent reinforcement learning approach,” IEEE Internet of Things Journal , 2025

work page 2025

Show all 15 references

[8]

Contract-based incentive mechanisms for uav networks in covert com- munications,

W. Tian, H. Fan, G. Liu, M. Du, Y . Liu, D. Niyato, and Z. Han, “Contract-based incentive mechanisms for uav networks in covert com- munications,” IEEE Transactions on V ehicular Technology, 2025

2025
[9]

Cross-modal generative semantic communications for mobile aigc: Joint semantic encoding and prompt engineering,

Y . Liu, H. Du, D. Niyato, J. Kang, Z. Xiong, S. Mao, P. Zhang, and X. Shen, “Cross-modal generative semantic communications for mobile aigc: Joint semantic encoding and prompt engineering,” IEEE Transactions on Mobile Computing , 2024

2024
[11]

Optimal lap altitude for maximum coverage,

A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal lap altitude for maximum coverage,” IEEE wireless communications letters , vol. 3, no. 6, pp. 569–572, 2014

2014
[12]

Optimized lte cell planning with varying spatial and temporal user densities,

H. Ghazzai, E. Yaacoub, M.-S. Alouini, Z. Dawy, and A. Abu-Dayya, “Optimized lte cell planning with varying spatial and temporal user densities,” IEEE Transactions on vehicular technology , vol. 65, no. 3, pp. 1575–1589, 2015

2015
[13]

3-d place- ment of an unmanned aerial vehicle base station (uav-bs) for energy- efficient maximal coverage,

M. Alzenad, A. El-Keyi, F. Lagum, and H. Yanikomeroglu, “3-d place- ment of an unmanned aerial vehicle base station (uav-bs) for energy- efficient maximal coverage,” IEEE Wireless Communications Letters , vol. 6, no. 4, pp. 434–437, 2017

2017
[14]

Satisfaction-aware incentive scheme for federated learning in industrial metaverse: Drl-based stackbelberg game approach,

X. Li, S. Qin, X. Tang, J. Kang, J. Ye, Z. Zhao, and D. Niyato, “Satisfaction-aware incentive scheme for federated learning in industrial metaverse: Drl-based stackbelberg game approach,” arXiv e-prints , pp. arXiv–2502, 2025

2025
[15]

R. S. Sutton, A. G. Barto et al., Reinforcement learning: An introduction. MIT press Cambridge, 1998, vol. 1, no. 1

1998

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [10]

Graph diffu- sion policy optimization,

Y . Liu, C. Du, T. Pang, C. Li, M. Lin, and W. Chen, “Graph diffu- sion policy optimization,” Advances in Neural Information Processing Systems, vol. 37, pp. 9585–9611, 2024

work page 2024

[2] [1]

Digital-twin- assisted task assignment in multi-uav systems: A deep reinforcement learning approach,

X. Tang, X. Li, R. Yu, Y . Wu, J. Ye, F. Tang, and Q. Chen, “Digital-twin- assisted task assignment in multi-uav systems: A deep reinforcement learning approach,” IEEE Internet of Things Journal , vol. 10, no. 17, pp. 15 362–15 375, 2023

work page 2023

[3] [2]

Uav-enabled communication strategy against detection in covert communication with asymmetric information,

W. Tian, J. Du, X. Ji, M. Du, G. Liu, and Z. Han, “Uav-enabled communication strategy against detection in covert communication with asymmetric information,” IEEE Transactions on Wireless Communica- tions, vol. 23, no. 7, pp. 6921–6937, 2023

work page 2023

[4] [3]

Dynamic topology organization and maintenance algorithms for autonomous uav swarms,

A. Gaydamaka, A. Samuylov, D. Moltchanov, M. Ashraf, B. Tan, and Y . Koucheryavy, “Dynamic topology organization and maintenance algorithms for autonomous uav swarms,” IEEE Transactions on Mobile Computing, vol. 23, no. 5, pp. 4423–4439, 2023

work page 2023

[5] [4]

Throughput maximiza- tion for covert uav relaying system,

H. Wang, H. Bai, F. Li, J. Chen, and L. Wang, “Throughput maximiza- tion for covert uav relaying system,” IEEE Transactions on V ehicular Technology, vol. 73, no. 3, pp. 4429–4434, 2023

work page 2023

[6] [5]

Generative ai based secure wireless sensing for isac networks. arxiv 2024,

J. Wang, H. Du, Y . Liu, G. Sun, D. Niyato, S. Mao, D. Kim, and X. Shen, “Generative ai based secure wireless sensing for isac networks. arxiv 2024,” arXiv preprint arXiv:2408.11398

arXiv 2024

[7] [6]

Covert communications: A comprehensive survey,

X. Chen, J. An, Z. Xiong, C. Xing, N. Zhao, F. R. Yu, and A. Nal- lanathan, “Covert communications: A comprehensive survey,” IEEE Communications Surveys & Tutorials , vol. 25, no. 2, pp. 1173–1198, 2023

work page 2023

[8] [7]

Dnn task assignment in uav networks: A generative ai enhanced multi-agent reinforcement learning approach,

X. Tang, Q. Chen, W. Weng, B. Liao, J. Wang, X. Cao, and X. Li, “Dnn task assignment in uav networks: A generative ai enhanced multi-agent reinforcement learning approach,” IEEE Internet of Things Journal , 2025

work page 2025