Pith. sign in

REVIEW 4 major objections 5 minor 43 references

Vision-Aided ISAC in Low-Altitude Economy Networks via De-Diffused Visual Priors

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that masked de-diffusion tokens can replace raw UAV imagery for ISAC resource scheduling, retaining enough semantic detail to reach within 4% of a raw-image upper bound.

desk verdict A genuinely novel architecture for privacy-preserving vision-aided ISAC, but the headline 4% claim is unsupported because the privacy-fidelity trade-off is never measured and the perception-efficiency equations are garbled. read the letter →

arxiv 2507.01574 v1 pith:FFHCFZHN submitted 2025-07-02 eess.SY cs.SY

Yulan Gao , Ziqiang Ye , Zhonghao Lyu , Ming Xiao , Yue Xiao , Ping Yang , Agata Manolova This is my paper
classification eess.SYcs.SY
keywords low-altitudeeconomynetworksintegratedsensingandcommunicationde-diffusionsemantictokensprivacy-preservingvisionDDPGresourceallocationRATselectionUAV-assisted
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 is trying to establish that a UAV can run its sensing-and-communication loop without ever uploading raw images. An onboard masked de-diffusion model compresses each frame into compact tokens describing agent type, activity, and heading; the cloud reconstructs synthetic images from those tokens and parses them with standard detectors, then fuses the result with mmWave radar measurements into a risk heatmap that guides radio access technology (RAT) selection, power control, and beamforming. The authors formulate this as a multi-objective optimization over energy and perception efficiency subject to user QoS constraints, and solve it with a DDPG-based policy called DeDiff-VARARO. Simulation results are the claim's main support: the token-driven policy converges reliably and stays within 4% of a raw-image upper bound on average reward, while transmitting only abstracted semantics. If this holds, UAV networks can cut visual overhead and privacy exposure while keeping near-raw-image scheduling quality.

What carries the argument

The load-bearing mechanism is the masked De-Diffusion pipeline plus cross-modal fusion. The operation $DeDiff(\cdot)$ maps a raw frame to structured text tokens $z^\text{text}_{n,t}$ that encode semantic type $sem_n$, activity class $act_n$, and heading $\theta^\text{vis}_{n,t}$ while masking sensitive content; a pretrained text-to-image diffusion model reconstructs synthetic imagery $\hat{I}^n_t$, and the object detector YOLOv11 plus action recognizer SlowFast parse it back into the semantic profile $z^\text{vis}_{n,t} \in \mathcal{S}$. These profiles are fused with FMCW radar outputs $(d_n, v_n, \psi_n)$ to form the risk heatmap $H_\text{vis}(x,y)$, which is fed together with SINR, channel, and radar states into a DDPG actor-critic that outputs RAT assignment $x_n$, transmit power $p_n$, and beamforming vector $w_n$. A multi-objective reward $r^n_t = \lambda_\text{EE}\,EE^n_t + \lambda_\text{PE}\,PE^n_t + \lambda_\text{SR}\,SR^n_t$ with a heavy penalty for QoS violation shapes the learning.

What would settle it

Collect real UAV footage over a dense urban block with known ground-truth positions and identities; run the full masked De-Diffusion-to-reconstruction-to-YOLOv11/SlowFast pipeline and measure both the average reward gap against the raw-image policy and the success of scene-recovery attacks on the reconstructed imagery. Finding either that reconstructed scenes expose recognizable landmarks or faces, or that the reward gap widens well beyond 4% under realistic occlusion and token corruption, would settle the claim.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the visual semantics needed for wireless resource control survive a privacy-preserving compression stage. A masked De-Diffusion model turns each raw image into textual tokens that keep coarse attributes such as agent type, activity class, and heading, while dropping spatially identifiable textures and landmarks. The server reconstructs a synthetic image with a pretrained diffusion model, YOLOv11 and SlowFast extract the semantic profile, radar supplies range, velocity, and angle, and the fused risk heatmap enters the state of a DDPG agent that chooses RAT assignment, transmit power, and beamforming. The numerical claim is a relative gap metric against a raw-image oracle, reported as within 4% average reward despite using only compact privacy-preserving tokens.

Load-bearing premise

The entire scheme assumes that masked De-Diffusion tokens are detailed enough that a pretrained text-to-image model can rebuild scenes that YOLOv11 and SlowFast parse correctly, yet abstract enough that no one can recover identifiable visual content from them.

Editorial extensions

If this is right

  • UAV access systems can cut visual uplink overhead by transmitting compact semantic tokens instead of raw frames while keeping near-oracle scheduling quality.
  • RAT selection and beam assignment can be conditioned on behavioral semantics such as heading and activity, not just SINR and position.
  • Fusing radar-derived motion estimates with visual risk heatmaps gives the learning policy a state representation that stays useful as agent count grows.
  • The multi-objective DDPG formulation can balance energy efficiency, perception efficiency, and link stability under per-agent QoS constraints.

Reading between the lines

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

  • If the 4% gap holds under realistic noise, the same token pipeline could replace raw imagery in other UAV control loops such as handover, collision avoidance, or trajectory planning, since they share the need for compact scene semantics.
  • The privacy claim is not yet tested against reconstruction attacks; a direct experiment could quantify how much spatial detail survives in the reconstructed images $\hat{I}^n_t$.
  • Token corruption, transmission loss, or adversarial semantic flips are untested; a small change in a heading or activity token could cascade into a different RAT or beam choice.
  • The raw-image upper bound is itself from simulation, so field deployment may widen the gap due to distribution shift between pretrained models and real low-altitude scenes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a vision-aided integrated sensing and communication (ISAC) framework for UAV-assisted low-altitude economy networks (LAENets). An onboard masked De-Diffusion model converts raw images into compact semantic tokens (agent type, activity class, heading); the cloud reconstructs synthetic images, parses them with YOLOv11 and SlowFast, fuses the results with mmWave radar measurements into a risk heatmap, and a DDPG agent selects RATs, transmit powers, and beamforming vectors. The central claim is that this privacy-preserving token-based pipeline achieves a time-averaged reward within 4% of a raw-image oracle while preserving user privacy and scaling to dense agent populations.

Significance. If substantiated, the paper would address a genuine and timely bottleneck: raw visual uploads from UAVs are bandwidth-hungry and privacy-sensitive, and a semantic-token interface that retains scheduling-relevant information would be valuable for ISAC in LAENets. The paper deserves credit for formulating a concrete cross-modal architecture, including an oracle upper bound and several baselines, and for identifying an important privacy-fidelity trade-off. However, the evidence presented does not currently establish that trade-off or the validity of the perception-efficiency objective, so the significance is conditional on substantial additional validation.

major comments (4)
  1. [Section IV-B, Eqs. (20)-(21)] The CRB expressions are internally inconsistent. Eq. (21) for CRBv contains the factor -As/(2σ2^2) without the exponential present in Eq. (20), so CRBv is negative for positive As, which is physically impossible for a variance bound. In addition, s_res is undefined, and σ2 is used for noise power in Eq. (8) while σ2^2 denotes perception-channel variance in Eqs. (20)-(21). Because the PE terms in Eqs. (18)-(19) and the reward in Eq. (28) depend on these CRBs, the simulation results cannot be interpreted until the expressions are corrected and properly sourced.
  2. [Section III-B and Section VI-C2] The central claim that DeDiff-VARARO achieves within 4% of the raw-image upper bound while preserving privacy is not supported by any measurement of either side of the trade-off. The paper reports no YOLOv11/SlowFast classification accuracy on reconstructed versus raw images, no heading-error distribution feeding Eq. (2), no ablation that corrupts or drops tokens, and no privacy metric such as image-retrieval success, LPIPS/SSIM similarity, or membership inference. The term 'semantic fidelity' used in the abstract and conclusion is never defined, and Eq. (34) compares only a time-averaged scalar reward, which could be dominated by SINR and energy terms rather than by the visual semantics that motivate the architecture. The headline result is therefore unverified.
  3. [Section IV-B, Eq. (22)] The optimization problem is stated as a maximization over w only, yet the objective contains p_n through the EE term, the constraints in (25) are on the RAT variables x_n, and the algorithm's action space in Eq. (27) includes x_n, p_n, and w_n. This mismatch makes the mathematical formulation ill-posed and disconnected from the DDPG action space. The decision variables should be explicitly stated as x, p, w, with the coupling among them reflected in the constraints.
  4. [Section VI-A and Algorithm 1] The simulation description says that MS-COCO and AVA v2.2 with pretrained YOLOv11 and SlowFast are used, but there is no description of how the masked De-Diffusion token extraction and text-to-image reconstruction are actually implemented or simulated. Without specifying how the De-Diffusion is trained or applied, and how reconstruction errors or privacy suppression are modeled, the reward curves in Figs. 5-7 cannot be reproduced or attributed to the proposed privacy-preserving pipeline.
minor comments (5)
  1. [Eq. (34)] The text says the gap is computed between the raw-image reward and the proposed method's reward, but the displayed formula uses r_ours while the surrounding text refers to r_dediff; please unify the notation.
  2. [References [42] and [43]] The references cited for Copilot and StableXL do not support the claim that these are text-to-image generators: [42] is about code completion with GitHub Copilot, and [43] is about adversarial attacks on Stable Diffusion. Please cite appropriate models or clarify the naming.
  3. [Section VI and Fig. 7] The caption of Fig. 7(b) says 'Average reward under different L', but the text describes performance versus the number of antennas M; please correct the mismatch.
  4. [Throughout] There are several typos and nomenclature inconsistencies, including 'sumbols', 'Cramero bound', 'rms bandwith', 'F ormulation', and the collision between σ2 (noise power) and σ2^2 (perception variance).
  5. [Fig. 2 caption] The caption mentions GPT4-assisted multi-stage visual data processing, but GPT-4 does not appear in the method description or the experiments; please clarify or remove this reference.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central 4%-of-oracle claim is anchored to a raw-image upper bound sharing the same reward, and the few self-citations are background or standard definitions.

full rationale

Walking the derivation chain, no step reduces to its own input by construction. Eq. (1) defines token extraction, Eq. (2) uses the heading prior for beam selection, Eqs. (3)-(5) give independent radar localization, Eqs. (6)-(13) model channels, Eqs. (14)-(16) perform risk-informed scheduling, Eqs. (22)-(25) state the optimization, and Eq. (28) defines the DDPG reward. The headline 'within 4%' claim is measured by Eq. (34), which compares time-averaged rewards against the Raw Image baseline; that baseline uses full-resolution visual input under the same DDPG framework and the same reward, so it is an external comparison anchor rather than a fitted constant renamed as a prediction. The reward in Eq. (28) is the training objective, so reward convergence is self-referential in the standard RL sense, but all baselines optimize the same reward, making the relative gaps meaningful. The paper's self-citations, e.g., [7], [15], and [39], are background or a standard definition of energy efficiency and are not load-bearing for the central claim; no uniqueness theorem is imported from the authors' prior work, and the De-Diffusion model [34], StableXL, YOLOv11, and SlowFast are external components. The unmeasured privacy-fidelity tradeoff and the undefined 'semantic fidelity' are validation gaps or correctness risks, not circular reductions under the definition used here.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim rests on unverified modeling assumptions: unproven and partly erroneous CRB formulas, an undefined risk heatmap, an untested assumption that reconstructed synthetic images retain semantics without leaking privacy, and unspecified balance thresholds. The free parameters listed are required for the reward and constraints but their values are not disclosed.

free parameters (5)
  • Reward balance parameters lambda_EE, lambda_PE, lambda_SR = not reported
    Weights in Eq. (28) set by hand with claimed sum constraint; values never disclosed in the simulation settings, so the objective actually optimized is not reproducible.
  • SINR threshold gamma_th = not reported
    Threshold in SR_n^t = I(gamma_n^t > gamma_th), Eq. (28), not specified.
  • RAT selection threshold delta_mm = not reported
    Threshold in Eq. (15) controlling mmWave assignment, not specified.
  • Minimum rate target R_min^{n,t} = not reported
    QoS constraint in Eq. (23), values not given.
  • Risk heatmap fusion weights = not reported
    H_vis(x,y) is described qualitatively; no equation, so any underlying parameters are undefined.
assumptions (6)
  • domain assumption LTE channel is flat Rayleigh fading (Eq. 10) with orthogonal resource allocation and no inter-user interference (Eq. 11).
    Assumed in Section III-D to make the LTE model tractable; standard but optimistic in dense LAENets.
  • domain assumption mmWave channel is a sparse geometric channel with L resolvable paths (Eq. 6).
    Standard geometric channel model in Section III-D; the paper does not validate against ray-tracing or measurements.
  • ad hoc to paper CRB expressions in Eqs. (20)-(21) are correct for ranging and speed estimation from pilot signals.
    No derivation or reference is given; the formulas as printed contain apparent sign and notation errors, so the perception efficiency objective is not well-grounded.
  • domain assumption Synthetic images reconstructed from semantic tokens preserve sufficient information for YOLOv11 and SlowFast to recover agent type, activity, and heading.
    Asserted in Section III-B and V-A; no fidelity experiment is reported.
  • domain assumption Semantic tokens suppress spatially identifiable content, ensuring privacy.
    Stated in contributions and footnote 1; no privacy metric, attack model, or leakage test is provided.
  • standard math Standard diffusion forward/reverse process definitions (Appendix A) hold.
    Standard DDPM equations, but the paper applies them to text-to-image reconstruction without specifying the actual sampler used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vision-Aided ISAC in Low-Altitude Economy Networks via De-Diffused Visual Priors." pith.science (2026). https://pith.science/paper/FFHCFZHN

@misc{pith2026250701574,
  author       = {Pith},
  title        = {Pith review of: Vision-Aided ISAC in Low-Altitude Economy Networks via De-Diffused Visual Priors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FFHCFZHN}},
  note         = {Machine review of arXiv:2507.01574}
}
abstract

Emerging low-altitude economy networks (LAENets) require agile and privacy-preserving resource control under dynamic agent mobility and limited infrastructure support. To meet these challenges, we propose a vision-aided integrated sensing and communication (ISAC) framework for UAV-assisted access systems, where onboard masked De-Diffusion models extract compact semantic tokens, including agent type, activity class, and heading orientation, while explicitly suppressing sensitive visual content. These tokens are fused with mmWave radar measurements to construct a semantic risk heatmap reflecting motion density, occlusion, and scene complexity, which guides access technology selection and resource scheduling. We formulate a multi-objective optimization problem to jointly maximize weighted energy and perception efficiency via radio access technology (RAT) assignment, power control, and beamforming, subject to agent-specific QoS constraints. To solve this, we develop De-Diffusion-driven vision-aided risk-aware resource optimization algorithm DeDiff-VARARO, a novel two-stage cross-modal control algorithm: the first stage reconstructs visual scenes from tokens via De-Diffusion model for semantic parsing, while the second stage employs a deep deterministic policy gradient (DDPG)-based policy to adapt RAT selection, power control, and beam assignment based on fused radar-visual states. Simulation results show that DeDiff-VARARO consistently outperforms baselines in reward convergence, link robustness, and semantic fidelity, achieving within $4\%$ of the performance of a raw-image upper bound while preserving user privacy and scalability in dense environments.

Figures

Figures reproduced from arXiv: 2507.01574 by the authors.

Figure 1
Figure 1. System architecture of the proposed vision-aided ISAC framework in LAENets. The architecture is composed of three tiers: (i) the ground layer, [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. De-Diffusion model and GPT4 Assisted multi-stage visual data [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. End-to-end pipeline for generating semantic risk-aware heatmaps from onboard visual inputs. The UAV first captures an input image, which is [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: (a)Convergence Validation, (b)Energy efficiency and perception quality over epochs. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Comparison of average reward over 100 time slots under different [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: (a)Average reward under different N, (b)Average reward under different L, and (c) Average reward under different pmax. REFERENCES [1] Z. Zhang, Y. Xiao, Z. Ma, M. Xiao, Z. Ding, X. Lei, G. K. Karagiannidis, and P. Fan, “6g wireless networks: Vision, requirements, archi…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 36 canonical work pages

  1. [1]

    6g wireless networks: Vision, requirements, architecture, and key technologies,

    Z. Zhang, Y . Xiao, Z. Ma, M. Xiao, Z. Ding, X. Lei, G. K. Karagiannidis, and P. Fan, “6g wireless networks: Vision, requirements, architecture, and key technologies,” IEEE vehicular technology magazine , vol. 14, no. 3, pp. 28–41, 2019

    work page 2019

  2. [2]

    What will 5g be?,

    J. G. Andrews, S. Buzzi, W. Choi, S. V . Hanly, A. Lozano, A. C. Soong, and J. C. Zhang, “What will 5g be?,” IEEE Journal on selected areas in communications , vol. 32, no. 6, pp. 1065–1082, 2014

    work page 2014

  3. [3]

    A survey on mobile edge computing: The communication perspective,

    Y . Mao, C. You, J. Zhang, K. Huang, and K. B. Letaief, “A survey on mobile edge computing: The communication perspective,” IEEE communications surveys & tutorials , vol. 19, no. 4, pp. 2322–2358, 2017

    2017

  4. [4]

    Deep sensing for future spectrum and location awareness 5g communications,

    B. Li, S. Li, A. Nallanathan, and C. Zhao, “Deep sensing for future spectrum and location awareness 5g communications,” IEEE Journal on Selected Areas in Communications , vol. 33, no. 7, pp. 1331–1344, 2015

    work page 2015

  5. [5]

    Integrated sensing and communication in 6g: Motivations, use cases, requirements, challenges and future directions,

    D. K. P. Tan, J. He, Y . Li, A. Bayesteh, Y . Chen, P. Zhu, and W. Tong, “Integrated sensing and communication in 6g: Motivations, use cases, requirements, challenges and future directions,” in 2021 1st IEEE International Online Symposium on Joint Communications & Sensing (JC&S), pp. 1–6, IEEE, 2021

    2021

  6. [6]

    Integration of communication and sensing in 6g: A joint industrial and academic perspective,

    H. Wymeersch, D. Shrestha, C. M. De Lima, V . Yajnanarayana, B. Richerzhagen, M. F. Keskin, K. Schindhelm, A. Ramirez, A. Wolf- gang, M. F. De Guzman, et al. , “Integration of communication and sensing in 6g: A joint industrial and academic perspective,” in 2021 IEEE 32nd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (...

    work page 2021

  7. [7]

    Joint maneuver and beamforming design for UA V-enabled integrated sensing and communication,

    Z. Lyu, G. Zhu, and J. Xu, “Joint maneuver and beamforming design for UA V-enabled integrated sensing and communication,” IEEE Trans- actions on Wireless Communications , vol. 22, no. 4, pp. 2424–2440, 2022

    2022

  8. [8]

    Integrated sensing and communication with mmwave massive MIMO: A compressed sampling perspective,

    Z. Gao, Z. Wan, D. Zheng, S. Tan, C. Masouros, D. W. K. Ng, and S. Chen, “Integrated sensing and communication with mmwave massive MIMO: A compressed sampling perspective,” IEEE Transactions on Wireless Communications, vol. 22, no. 3, pp. 1745–1762, 2022

    work page 2022

Show all 43 references
  1. [9]

    Spectral co- existence of mimo radar and mimo cellular system,

    J. A. Mahal, A. Khawar, A. Abdelhadi, and T. C. Clancy, “Spectral co- existence of mimo radar and mimo cellular system,” IEEE Transactions on Aerospace and Electronic Systems , vol. 53, no. 2, pp. 655–668, 2017

    2017

  2. [10]

    Phase- modulation based dual-function radar-communications,

    A. Hassanien, M. G. Amin, Y . D. Zhang, and F. Ahmad, “Phase- modulation based dual-function radar-communications,” IET Radar , Sonar & Navigation , vol. 10, no. 8, pp. 1411–1421, 2016

    2016

  3. [11]

    Semantic-aware vision-assisted integrated sensing and communication: Architecture and resource allocation,

    Y . Lu, W. Mao, H. Du, O. A. Dobre, D. Niyato, and Z. Ding, “Semantic-aware vision-assisted integrated sensing and communication: Architecture and resource allocation,” IEEE Wireless Communications , vol. 31, no. 3, pp. 302–308, 2024

    2024

  4. [12]

    Integrated sensing, computing and semantic communication for vehicular networks,

    Y . Yang, Z. Yang, C. Huang, W. Xu, Z. Zhang, D. Niyato, and M. Shikh- Bahaei, “Integrated sensing, computing and semantic communication for vehicular networks,” IEEE Transactions on V ehicular Technology, 2025

    2025

  5. [13]

    3d scene-based beam selection for mmwave communications,

    W. Xu, F. Gao, S. Jin, and A. Alkhateeb, “3d scene-based beam selection for mmwave communications,” IEEE Wireless Communications Letters , vol. 9, no. 11, pp. 1850–1854, 2020

    2020

  6. [14]

    A survey on fundamental limits of integrated sensing and communication,

    A. Liu, Z. Huang, M. Li, Y . Wan, W. Li, T. X. Han, C. Liu, R. Du, D. K. P. Tan, J. Lu, et al. , “A survey on fundamental limits of integrated sensing and communication,” IEEE Communications Surveys & Tutorials, vol. 24, no. 2, pp. 994–1034, 2022

    2022

  7. [15]

    Space-air-ground integrated wireless networks for 6g: Basics, key technologies and future trends,

    Y . Xiao, Z. Ye, M. Wu, H. Li, M. Xiao, M.-S. Alouini, A. Al-Hourani, and S. Cioni, “Space-air-ground integrated wireless networks for 6g: Basics, key technologies and future trends,” IEEE Journal on Selected Areas in Communications , 2024

    2024

  8. [16]

    Optimal transmit beamforming for integrated sensing and communication,

    H. Hua, J. Xu, and T. X. Han, “Optimal transmit beamforming for integrated sensing and communication,” IEEE Transactions on V ehicular Technology, 2023

    2023

  9. [17]

    Joint design of communication and sensing for beyond 5g and 6g systems,

    T. Wild, V . Braun, and H. Viswanathan, “Joint design of communication and sensing for beyond 5g and 6g systems,” IEEE Access , vol. 9, pp. 30845–30857, 2021

    2021

  10. [18]

    Integrated sensing and communication for low altitude econ- omy: Opportunities and challenges,

    Y . Jiang, X. Li, G. Zhu, H. Li, J. Deng, K. Han, C. Shen, Q. Shi, and R. Zhang, “Integrated sensing and communication for low altitude econ- omy: Opportunities and challenges,” IEEE Communications Magazine , 2025

    2025

  11. [19]

    Networked isac for low-altitude economy: Coordinated transmit beamforming and UA V trajectory de- sign,

    G. Cheng, X. Song, Z. Lyu, and J. Xu, “Networked isac for low-altitude economy: Coordinated transmit beamforming and UA V trajectory de- sign,” IEEE Transactions on Communications , 2025

    2025

  12. [20]

    Co- operative ISAC-empowered low-altitude economy,

    J. Tang, Y . Yu, C. Pan, H. Ren, D. Wang, J. Wang, and X. You, “Co- operative ISAC-empowered low-altitude economy,” IEEE Transactions on Wireless Communications , 2025

    2025

  13. [21]

    Networked ISAC based UA V tracking and handover towards low-altitude economy,

    Y . Feng, C. Zhao, H. Luo, F. Gao, F. Liu, and S. Jin, “Networked ISAC based UA V tracking and handover towards low-altitude economy,”IEEE Transactions on Wireless Communications , 2025

    2025

  14. [22]

    Integrated sensing and communications for low-altitude economy: A deep reinforcement learning approach,

    X. Ye, Y . Mao, X. Yu, S. Sun, L. Fu, and J. Xu, “Integrated sensing and communications for low-altitude economy: A deep reinforcement learning approach,” arXiv preprint arXiv:2412.04074 , 2024

    2024 arXiv

  15. [23]

    Computer vision aided mmwave beam alignment in V2X communications,

    W. Xu, F. Gao, X. Tao, J. Zhang, and A. Alkhateeb, “Computer vision aided mmwave beam alignment in V2X communications,” IEEE Transactions on Wireless Communications , vol. 22, no. 4, pp. 2699– 2714, 2022

    2022

  16. [24]

    Millimeter wave base stations with cameras: Vision-aided beam and blockage prediction,

    M. Alrabeiah, A. Hredzak, and A. Alkhateeb, “Millimeter wave base stations with cameras: Vision-aided beam and blockage prediction,” in 2020 IEEE 91st vehicular technology conference (VTC2020-Spring) , pp. 1–5, IEEE, 2020

    2020

  17. [25]

    From ground to sky: Architectures, applications, and challenges shaping low-altitude wireless networks,

    W. Yuan, Y . Cui, J. Wang, F. Liu, G. Sun, T. Xiang, J. Xu, S. Jin, D. Niyato, S. Coleri, et al. , “From ground to sky: Architectures, applications, and challenges shaping low-altitude wireless networks,” arXiv preprint arXiv:2506.12308 , 2025

    2025 arXiv

  18. [26]

    Joint transmit beamforming for multiuser mimo communications and mimo radar,

    X. Liu, T. Huang, N. Shlezinger, Y . Liu, J. Zhou, and Y . C. Eldar, “Joint transmit beamforming for multiuser mimo communications and mimo radar,” IEEE Transactions on Signal Processing, vol. 68, pp. 3929–3944, 2020

    2020

  19. [27]

    SIGN-FCF: Sign-based federated collaborative filtering for privacy-preserving personalized rec- ommendation,

    Z. Zhou, L. Xu, L. Zhu, K. Gai, and P. Jiang, “SIGN-FCF: Sign-based federated collaborative filtering for privacy-preserving personalized rec- ommendation,” in 2025 IEEE 10th International Conference on Smart Cloud (SmartCloud) , pp. 50–55, IEEE, 2025

    2025

  20. [28]

    Multiobjective reinforcement learning: A comprehensive overview,

    C. Liu, X. Xu, and D. Hu, “Multiobjective reinforcement learning: A comprehensive overview,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , vol. 45, no. 3, pp. 385–398, 2014

    2014

  21. [29]

    Adaptive control and intersections with reinforce- ment learning,

    A. M. Annaswamy, “Adaptive control and intersections with reinforce- ment learning,” Annual Review of Control, Robotics, and Autonomous Systems, vol. 6, no. 1, pp. 65–93, 2023. 13

    2023

  22. [30]

    Temporal spectrum cartography in low- altitude economy networks: A generative ai framework with multi-agent learning,

    C. Zhao, R. Zhang, J. Wang, D. Niyato, G. Sun, H. Du, Z. Li, A. Jamalipour, and D. I. Kim, “Temporal spectrum cartography in low- altitude economy networks: A generative ai framework with multi-agent learning,” arXiv preprint arXiv:2505.15571 , 2025

    2025 arXiv

  23. [31]

    Learning in the air: Secure federated learning for UA V-assisted crowdsensing,

    Y . Wang, Z. Su, N. Zhang, and A. Benslimane, “Learning in the air: Secure federated learning for UA V-assisted crowdsensing,”IEEE Trans- actions on network science and engineering , vol. 8, no. 2, pp. 1055– 1069, 2020

    2020

  24. [32]

    Pushing AI to wireless network edge: An overview on integrated sensing, communication, and computation towards 6G,

    G. Zhu, Z. Lyu, X. Jiao, P. Liu, M. Chen, J. Xu, S. Cui, and P. Zhang, “Pushing AI to wireless network edge: An overview on integrated sensing, communication, and computation towards 6G,” Science China Information Sciences , vol. 66, no. 3, p. 130301, 2023

    2023

  25. [33]

    Privacy- preserving federated learning for UA V-enabled networks: Learning- based joint scheduling and resource management,

    H. Yang, J. Zhao, Z. Xiong, K.-Y . Lam, S. Sun, and L. Xiao, “Privacy- preserving federated learning for UA V-enabled networks: Learning- based joint scheduling and resource management,” IEEE Journal on Selected Areas in Communications , vol. 39, no. 10, pp. 3144–3159, 2021

    2021

  26. [34]

    De-diffusion makes text a strong cross-modal interface,

    C. Wei, C. Liu, S. Qiao, Z. Zhang, A. Yuille, and J. Yu, “De-diffusion makes text a strong cross-modal interface,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 13492–13503, 2024

    2024

  27. [35]

    Diffusion models in vision: A survey,

    F.-A. Croitoru, V . Hondru, R. T. Ionescu, and M. Shah, “Diffusion models in vision: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 9, pp. 10850–10869, 2023

    2023

  28. [36]

    Ultralytics yolo11

    “Ultralytics yolo11.” https://docs.ultralytics.com/models/yolo11/, 2025

    2025

  29. [37]

    Pyslowfast

    H. Fan, Y . Li, B. Xiong, W.-Y . Lo, and C. Feichtenhofer, “Pyslowfast.” https://github.com/facebookresearch/slowfast, 2020

    2020

  30. [38]

    Energy-efficient scheduling and power allocation in downlink OFDMA networks with base station coordination,

    L. Venturino, A. Zappone, C. Risi, and S. Buzzi, “Energy-efficient scheduling and power allocation in downlink OFDMA networks with base station coordination,” IEEE transactions on wireless communica- tions, vol. 14, no. 1, pp. 1–14, 2014

    2014

  31. [39]

    Dynamic social- aware peer selection for cooperative relay management with D2D communications,

    Y . Gao, Y . Xiao, M. Wu, M. Xiao, and J. Shao, “Dynamic social- aware peer selection for cooperative relay management with D2D communications,” IEEE Transactions on Communications, vol. 67, no. 5, pp. 3124–3139, 2019

    2019

  32. [40]

    Coco: Common objects in context

    “Coco: Common objects in context.” https://cocodataset.org/#home, 2021

    2021

  33. [41]

    Ava: A video dataset of spatio-temporally localized atomic visual actions,

    C. Gu, C. Sun, D. A. Ross, C. V ondrick, C. Pantofaru, Y . Li, S. Vi- jayanarasimhan, G. Toderici, S. Ricco, R. Sukthankar, et al. , “Ava: A video dataset of spatio-temporally localized atomic visual actions,” in Proceedings of the IEEE conference on computer vision and patter...

    2018

  34. [42]

    Github copilot ai pair programmer: Asset or liability?,

    A. M. Dakhel, V . Majdinasab, A. Nikanjam, F. Khomh, M. C. Desmarais, and Z. M. J. Jiang, “Github copilot ai pair programmer: Asset or liability?,” Journal of Systems and Software , vol. 203, p. 111734, 2023

    2023

  35. [43]

    A pilot study of query-free adversarial attack against stable diffusion,

    H. Zhuang, Y . Zhang, and S. Liu, “A pilot study of query-free adversarial attack against stable diffusion,” in Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition , pp. 2384–2391, 2023. APPENDIX A DIFFUSION PROCESS We denote the original data a...

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.