Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Chaining 802.11 deauthentication with stolen credentials turns a brief wireless blackout into undetected, persistent participation in drone federated learning.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 10:16 UTC pith:WZFYQMN7

load-bearing objection Solid conditional chain attack: deauth + credential reuse opens Flower, but 'practical' overstates it when credential extraction is assumed, not demonstrated. the 3 major comments →

arxiv 2607.20280 v1 pith:WZFYQMN7 submitted 2026-07-22 cs.CR

Chained Attacks on Drone-Based Federated Learning: From Network Disruption to Device Impersonation

Suleiman Muhammad Sabo , Hamed Alkharsh , Peilin Li , Chuadhry Mujeeb Ahmed , Aydin Abadi , Shishir Nagaraja , Rajiv Ranjan This is my paper
classification cs.CR
keywords Federated learningDrone swarmsDeauthentication attackDevice impersonationSingle-factor authenticationNon-IID data802.11Edge intelligence
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that two well-known primitives—forged 802.11 disconnect frames and replayed node credentials—can be chained into one practical attack on federated learning in drone swarms. It shows on two physical edge-computer testbeds that forcing clients offline degrades model accuracy, with the largest losses under label-skewed (non-IID) data, and that a server using single-factor authentication accepts the impersonator as the returning legitimate node with no security alert. The deeper point is that availability and identity are not independent: a temporary denial-of-service becomes a persistent integrity breach once the attacker holds one node's credentials.

Core claim

On the paper's own terms, the discovery is that stateless federated aggregation combined with possession-based authentication gives an attacker a seamless replacement path: after a deauthentication attack forces a drone offline, its credentials alone—no physical airframe—are enough to rejoin and contribute updates, and the server logs the event as routine churn. Under non-IID data the same attack becomes selective: disconnecting 'expert' nodes that hold concentrated class knowledge produces disproportionate, class-specific accuracy loss and delays convergence, a targeting effect that a generic wireless denial of service cannot achieve. The authors further report that concurrent credential re

What carries the argument

The load-bearing mechanism is the chained pairing of the unauthenticated 802.11 deauthentication frame—a forged management packet that any nearby radio can send to force a station off the network—with the federated server's single-factor, possession-based node authentication. The deauthentication opens a vacancy; the credential replay fills it. The chain's leverage comes from the aggregator being stateless and built to tolerate client churn: a node that drops and returns with matching credentials is treated as legitimate. Non-IID data amplify the effect, because certain clients hold concentrated expertise whose removal causes class-specific degradation.

Load-bearing premise

The attack chain collapses to ordinary denial-of-service if the adversary does not already hold a valid credential for a target drone—an acquisition step the paper assumes rather than demonstrates—and the testbed results likewise assume that single-board computers faithfully represent real aircraft at the wireless and authentication layers.

What would settle it

Run the two-stage attack on an identical federation with 802.11w (management frame protection) enabled on the access point and with an impersonator that uses no stolen private key, only the victim's MAC address. If the deauthentication burst fails to force clients offline, or the credential-less client is rejected, the chain's claimed practicality is disproven in that configuration.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A one-round disconnect window (about 58 seconds in the experiments) is enough for an impersonator to join the federation and contribute updates before anything looks wrong.
  • Under non-IID data, an attacker who knows the label distribution can target expert nodes and induce selective, class-specific blindness while aggregate accuracy still appears acceptable.
  • Because the aggregator is stateless, the attack leaves no application-layer alert; detection requires correlating wireless-layer events with training logs.
  • If credentials are static and reused across missions, one key extraction permits indefinite impersonation even after the legitimate drone is recovered or destroyed.
  • Defenses aimed only at network availability or only at authentication are insufficient; the paper argues for combined measures including hardware attestation and freshness proofs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same vacancy-exploitation logic applies to any churn-tolerant federated deployment with static per-node credentials, not just drones; the drone setting mainly adds physical capture as a realistic key-leak path.
  • The impersonation step could be weaponized beyond silent participation: a legitimate-looking identity is exactly the cover needed to inject poisoned updates, converting the chain into a credentialed Byzantine attack.
  • A minimal defense that the paper does not develop is session-continuity tracking—requiring a client to prove it held the connection through the round, or binding round participation to a monotonic freshness counter—which would close the replay window without changing the aggregation algorithm.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents a two-stage chained attack against drone-based federated learning: (1) an 802.11 deauthentication attack that forces legitimate clients offline, and (2) credential-based impersonation of the disconnected client. The authors validate Stage 1 on physical Raspberry Pi and NVIDIA Jetson testbeds running Flower/FedAvg with CIFAR-10 under IID and non-IID partitions, quantifying accuracy degradation, convergence instability, and connectivity cost. They validate Stage 2 on a five-client federation, showing that the Flower server accepts a pre-provisioned impersonator after an adversarial disconnection while rejecting concurrent credential reuse. The paper argues that single-factor possession-based authentication combined with stateless FL aggregation leaves a persistent vulnerability window after any disconnect, and that under non-IID data the loss of 'expert' clients causes disproportionate, class-specific degradation. The authors are transparent about several boundary conditions, including the assumption that the adversary already possesses valid credentials and the use of compute surrogates rather than airborne drones.

Significance. If the claims hold, the paper makes a useful empirical contribution: it demonstrates on real hardware that a commodity Wi-Fi deauthentication attack can create the vacancy needed for credential impersonation in a stateful-authentication/stateless-aggregation FL framework, and it quantifies the interaction between availability attacks and non-IID data skew. The strongest parts are the direct server-log evidence for Stage 2 and the explicit acknowledgement that the credential-possession precondition is a threat-model assumption rather than a demonstrated capability. The attack chain itself—deauth, disconnect, clone, rejoin—is observed end-to-end in the Stage 2 timeline. The paper does not rely on fitted parameters or circular derivations; the main quantitative contribution is reproducible in principle, though no code is shipped. The significance is moderate: the individual primitives (deauth and credential replay) are well-known, and the paper's contribution is their combination and hardware-validated measurement, not a new vulnerability class. The practical-impact claim is currently stronger than the evidence supports, because credential acquisition and extraction are not demonstrat

major comments (3)
  1. [Abstract; §IV-C; §V-B] The central claim of a 'practical chained attack' using 'extracted credentials' is only demonstrated conditional on the adversary already possessing valid credentials. §IV-C explicitly states 'It is assumed that the adversary has obtained valid client credentials through any available means', and §V-B pre-loads the impersonator with credentials, so the experiment demonstrates replay, not extraction. The listed acquisition paths (physical capture, supply chain, insider, insecure provisioning) are not measured or demonstrated on the target devices. If real drone identity material is held in non-exportable secure storage or bound to hardware, Stage 2 fails and the chain reduces to ordinary DoS. The authors are transparent about this boundary, but the abstract and conclusion wording overstates what is shown. Either add a credential-extraction/replay demonstration on representative hardware o
  2. [§V-A; Tables II–III] All accuracy numbers are single runs with no variance estimates. The main quantitative findings—non-IID degradation 'more than double' IID on Pi (15.75 vs 7.20), Jetson's faster convergence partially offsetting loss, and the expert-node '3-round delay'—are point estimates from one seed and one Dirichlet partition. With 10 rounds and stochastic training, these differences could lie within run-to-run noise. The paper's own future-work section (§VIII) calls for multi-trial validation; before publication the central quantitative claims need at least repeated seeds (e.g., 3–5) and reported intervals, or the claims should be softened to single-run observations.
  3. [§V-A1 vs Table II] There is a numeric inconsistency in the IID Raspberry Pi result. §V-A1 says the attacked system 'converged to 49.80%' at round 10, a 7.20% loss, and that the attack ran 'rounds 5 to 8 at escalating intensity.' Table II reports round-8 IID degradation of 6.40% and round-7 of 7.20%, so the final 7.20% gap is not the round-8 value. Clarify whether the attack is cumulative or per-round, and make the reported final accuracy consistent with the table.
minor comments (5)
  1. [§IV-B4 vs §V-A5] The metric C_C is defined as 'per-round latency penalty from reconnection' but is computed as the difference in mean accuracy between baseline and attack during the attack window. Please align the definition and the calculation.
  2. [§V-B] The text says the impersonator was 'pre-configured with credentials extracted from Client 2 during Stage 1,' but Stage 1 is a deauthentication attack and no credential-extraction procedure is described. Say 'credentials obtained from a prior compromise' or describe the extraction method.
  3. [Fig. 5b] The figure shows intensity labels 20%, 40%, 60% but the caption does not explain how these map to rounds. Add the round-to-intensity mapping or a legend.
  4. [Table I] For this work, the 'Physical Capture' column is marked 'Yes,' but no physical-capture or on-device extraction experiment is performed. Mark it as 'Assumed' or 'Threat model' to avoid overstating the contribution.
  5. [§IV-B2] The plaintext packet-inspection phase is described as 'essential context for subsequent analysis of malicious model updates,' but no malicious-update analysis is presented. Either connect this to a later result or trim the discussion.

Circularity Check

0 steps flagged

No significant circularity: the attack chain is empirically tested, and the credential-possession precondition is an explicit threat-model assumption, not a fitted input or derived result.

full rationale

The paper's derivation chain is empirical rather than derivational. Stage 1 measures accuracy degradation under 802.11 deauthentication on two physical testbeds; these are observed outcomes, not quantities fitted to the claimed conclusion. Stage 2 directly observes Flower server logs during impersonation and reports that the server 'accepted the new node, assigned an identifier, and served the global model' with 'no authentication failure ... logged'. No equation in the paper reduces a predicted quantity to a fitted parameter. The credential-possession precondition is explicitly stated as a threat-model assumption: 'It is assumed that the adversary has obtained valid client credentials through any available means ... The focus is on assessing the system's resilience to impersonation, rather than the method of credential acquisition.' This is a boundary condition, not a circular step. Similarly, the PMF-disabled wireless configuration and the compute-surrogate testbeds are stated limitations, not hidden inputs. The only self-citation ([3]) is a background reference on federated edge intelligence and is not load-bearing for any central claim. There is no imported uniqueness theorem, no ansatz smuggled in via citation, and no renaming of a known result as a new derivation. The paper is self-contained against concrete benchmarks and transparent about its assumptions, so the honest finding is no significant circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The paper is empirical, not derivational; it relies on threat-model preconditions and platform-transfer assumptions rather than postulating new physics or mechanisms. The quantitative outputs depend on the listed configuration choices.

free parameters (4)
  • Dirichlet concentration α = 0.3
    Hand-chosen to create non-IID label skew; directly controls the data heterogeneity that drives the measured vulnerability.
  • Number of global rounds = 10
    Chosen by the authors because further rounds gave marginal improvement; all accuracy baselines depend on this cutoff.
  • Attack intensity schedule = 20/40/60/80% (rounds 5-8 Pi; 20/40/60% rounds 5-7 Jetson)
    Imposed by the authors; the ΔA figures are measured against this schedule, and the Jetson sweep was truncated because higher intensities gave marginal extra degradation.
  • Dirichlet random seed = 42
    Single seed for partitioning; no repeated-seed variance is reported.
axioms (4)
  • domain assumption 802.11 management frames are unencrypted and PMF/802.11w is disabled on target drone networks.
    The testbed network runs without PMF (Section IV-A); the authors acknowledge this as a boundary condition and cite inconsistent PMF support on IoT platforms.
  • domain assumption The adversary possesses valid client credentials for at least one target node.
    Stated in Section III-A as a precondition; acquisition via physical capture, insider, or supply chain is not demonstrated or measured.
  • domain assumption Raspberry Pi and Jetson compute surrogates reproduce the drone attack surface at the 802.11 link and FL authentication layers.
    Section IV argues airframe dynamics are irrelevant to these layers; no airborne validation is provided.
  • domain assumption Flower's single-session/possession-based authentication is representative of general FL frameworks.
    Section VII-B generalizes a conceptual flaw beyond Flower, but only Flower is tested.

pith-pipeline@v1.3.0-alltime-deepseek · 11149 in / 12468 out tokens · 108410 ms · 2026-08-01T10:16:05.312698+00:00 · methodology

0 comments
read the original abstract

Edge Intelligence (EI) has emerged as a transformative model for mission-critical unmanned platforms, such as drone swarms, by enabling collaborative model training at the network periphery. However, the security of FL deployments depends on both network availability and robust client authentication mechanisms. This paper investigates a chained attack against drone-based FL systems that combines network-layer denial-of-service with credential-based impersonation. We demonstrate that an adversary can: (1) force legitimate drones offline using 802.11 deauthentication attacks, and (2) subsequently impersonate the disconnected drone using extracted credentials. Through a systematic literature review and empirical validation using the Flower framework on two distinct testbeds of Raspberry Pi and Jetsons, we quantify the impact of availability disruptions under Independent and Identically Distributed (IID) and Non-Independently and Identically Distributed (Non-IID) data distributions, and confirm that single-factor authentication permits post-disconnect impersonation. Our findings reveal that even short-term wireless interruptions cascade into substantial training instability, particularly under non-IID conditions, while the authentication gap enables adversaries to seamlessly replace disconnected nodes. We discuss the compounded implications for mission-critical drone deployments and outline directions for future defenses addressing both availability and authentication vulnerabilities.

Figures

Figures reproduced from arXiv: 2607.20280 by Aydin Abadi, Chuadhry Mujeeb Ahmed, Hamed Alkharsh, Peilin Li, Rajiv Ranjan, Shishir Nagaraja, Suleiman Muhammad Sabo.

Figure 1
Figure 1. Figure 1: Targeted reconnaissance of the swarm BSSID, mapping [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Chained attack workflow: Stage 1 forces disconnection [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Experimental testbed: 8-node Jetson cluster with ded [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Wireshark capture of the Flower PushObjectRequest packet. Protocol Buffers. A custom Python decoder was implemented to interpret the payload. The decoding process identified a fully connected layer with shape (10, 84). An excerpt of the recovered weight matrix is presented below: Preview of weights: [[ 1.83726791e-02 -2.09797233e-01 4.93503697e-02 -2.98543628e-02 8.38919808e-02 6.39304482e-02 -5.98880658e-… view at source ↗
Figure 2
Figure 2. Figure 2: V. RESULTS A. Stage 1: Impact of Network Disruption 1) IID Data Distribution (Raspberry Pi vs. Jetson) Under IID conditions, both hardware platforms demon￾strated resilience to availability attacks, though with distinct absolute accuracy levels and degradation patterns. The Rasp￾berry Pi testbed achieved a baseline accuracy of 57.00% after 10 training rounds, which declined to 49.80% under sustained attack… view at source ↗
Figure 5
Figure 5. Figure 5: Accuracy trajectories across training rounds under IID and Non-IID conditions for both testbeds. The red shaded regions [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Symbolic Attack Chain Generation from Atomic Red Team Techniques: An Empirical Study of Predicate Representation Granularity

    cs.CR 2026-07 reject novelty 5.0

    A 16-technique ablation of attack-planning predicate granularity reports invariance that is guaranteed by byte-identical relabeling, with only one technique showing a genuine resolution gain.

Reference graph

Works this paper leans on

34 extracted references · 3 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Distributed collaborative data processing framework for unmanned platforms based on federated edge intelligence,

    S. Liu, N. Shan, X. Bao, and X. Xu, “Distributed collaborative data processing framework for unmanned platforms based on federated edge intelligence,”Sensors, vol. 25, no. 15, p. 4752, 2025

    2025

  2. [2]

    Communication-efficient learning of deep networks from de- centralized data,

    H. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. Ag ¨uera y Arcas, “Communication-efficient learning of deep networks from de- centralized data,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), ser. Proceedings of Machine Learning Research, vol. 54. PMLR, 2017, pp. 1273–1282

    2017

  3. [3]

    Decentralized weather forecasting via dis- tributed machine learning and blockchain-based model validation,

    A. Abadi, B. Aldali, B. Vincent, E. A. J. Hurley, H. Aljazaeri, J. Hedley- Cook, J.-L. Bell, L. Uwuigbusun, M. Ahmed, R. Umar, S. Nagaraja, S. Sabo, and W. Alrbeiqi, “Decentralized weather forecasting via dis- tributed machine learning and blockchain-based model validation,” in 2025 3rd International Conference on Federated Learning Technologies and Appli...

    2025

  4. [4]

    Federated learning for internet of things: A comprehensive survey,

    D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, and H. V . Poor, “Federated learning for internet of things: A comprehensive survey,”IEEE Communications Surveys & Tutorials, vol. 23, no. 3, pp. 1622–1658, 2021

    2021

  5. [5]

    Intrusion detection in 5G and Wi-Fi networks: A survey of current methods, challenges, and perspectives,

    C. Hamroun, A. Fladenmuller, M. Pariente, and G. Pujolle, “Intrusion detection in 5G and Wi-Fi networks: A survey of current methods, challenges, and perspectives,”IEEE Access, vol. 13, pp. 1–34, 2025, early Access

    2025

  6. [6]

    A survey on security and privacy of federated learning,

    V . Mothukuri, R. M. Parizi, S. Pouriyeh, Y . Huang, A. Dehghantanha, and G. Srivastava, “A survey on security and privacy of federated learning,”Future Generation Computer Systems, vol. 115, pp. 619–640, 2021

    2021

  7. [7]

    Flower: A friendly federated learning research framework,

    D. J. Beutel, T. Topal, A. Mathur, X. Qiu, J. Fernandez-Marques, Y . Gao, L. Sani, H. K. Li, T. Parcollet, P. P. B. de Gusm ˜ao, and N. D. Lane, “Flower: A friendly federated learning research framework,”arXiv preprint arXiv:2007.14390, 2020

    Pith/arXiv arXiv 2007

  8. [8]

    Securing UA V communication: Authenti- cation and integrity,

    M. Ouadah and F. Merazka, “Securing UA V communication: Authenti- cation and integrity,” in2024 11th International Conference on Wireless Networks and Mobile Communications (WINCOM). IEEE, 2024

    2024

  9. [9]

    SELTHA: Secure, efficient and lightweight authentication mechanism for unmanned aerial vehicle network,

    W. Salam, S. K.-u.-R. Raazi, and N. H. Ansari, “SELTHA: Secure, efficient and lightweight authentication mechanism for unmanned aerial vehicle network,” in2023 7th International Multi-Topic ICT Conference (IMTIC). IEEE, 2023, pp. 1–7

    2023

  10. [10]

    Secure blockchain-enabled authentication key management framework with big data analytics for drones in networks beyond 5G applications,

    A. K. Mishra, M. Wazid, D. P. Singh, A. K. Das, J. Singh, and A. V . Vasilakos, “Secure blockchain-enabled authentication key management framework with big data analytics for drones in networks beyond 5G applications,”Drones, vol. 7, no. 8, p. 508, 2023

    2023

  11. [11]

    BC- IoDT: Blockchain-based framework for authentication in internet of drone things,

    J. Akram, A. Akram, R. H. Jhaveri, M. Alazab, and H. Chi, “BC- IoDT: Blockchain-based framework for authentication in internet of drone things,” inProceedings of the 5th International ACM Mobicom Workshop on Drone Assisted Wireless Communications for 5G and Beyond. ACM, 2022, pp. 115–120

    2022

  12. [12]

    Au- thentication and handover challenges and methods for drone swarms,

    Y . Aydin, G. Karabulut-Kurt, E. Ozdemir, and H. Yanikomeroglu, “Au- thentication and handover challenges and methods for drone swarms,” CoRR, vol. abs/2201.05657, 2022

    Pith/arXiv arXiv 2022

  13. [13]

    Scalable authenticated communication in drone swarm environment,

    K. Han, E. Al Nuaimi, S. Al Blooshi, R. Psiakis, and C. Y . Yeun, “Scalable authenticated communication in drone swarm environment,” Journal of Internet Technology, 2024

    2024

  14. [14]

    FPGA- based dual-layer authentication scheme utilizing AES and ECC for un- manned aerial vehicles,

    K. S. Roy, M. Sujith, B. Bhanu, Preethi, and R. A. Hazarika, “FPGA- based dual-layer authentication scheme utilizing AES and ECC for un- manned aerial vehicles,”EURASIP Journal on Wireless Communications and Networking, vol. 2024, no. 1, p. 91, 2024

    2024

  15. [15]

    A novel lightweight authentication mechanism for uavs based on sddn architecture,

    N. Kammoun, A. B. C. Douss, and R. Abassi, “A novel lightweight authentication mechanism for uavs based on sddn architecture,” in 2024 20th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob). IEEE, 2024, pp. 1–6

    2024

  16. [16]

    Efficient remote identification for drone swarms,

    K.-M. Seo, J. Kim, S. Lee, J.-W. Kwon, and S.-H. Seo, “Efficient remote identification for drone swarms,”Computers, Materials and Continua, 2023

    2023

  17. [17]

    A privacy-preserving access control protocol for 6G supported intelligent UA V networks,

    K. Mahmood, S. Shamshad, M. H. Anisi, A. Brighente, M. A. Saleem, and A. K. Das, “A privacy-preserving access control protocol for 6G supported intelligent UA V networks,”Vehicular Communications, vol. 54, p. 100937, 2025

    2025

  18. [18]

    A sur- vey on security of UA V swarm networks: Attacks and countermeasures,

    X. Wang, Z. Zhao, L. Yi, Z. Ning, L. Guo, F. R. Yu, and S. Guo, “A sur- vey on security of UA V swarm networks: Attacks and countermeasures,” ACM Computing Surveys, vol. 57, no. 3, pp. 74:1–74:37, 2024

    2024

  19. [19]

    Advances and open problems in federated learning,

    P. Kairouz, H. B. McMahanet al., “Advances and open problems in federated learning,”Foundations and Trends in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021

    2021

  20. [20]

    LAFED: A lightweight authentication mechanism for blockchain-enabled federated learning system,

    S. Ji, J. Zhang, Y . Zhang, Z. Han, and C. Ma, “LAFED: A lightweight authentication mechanism for blockchain-enabled federated learning system,”Future Generation Computer Systems, vol. 145, pp. 56–67, 2023

    2023

  21. [21]

    Efficient and privacy-preserving authentication for federated learning in industrial internet of things data sharing application,

    Y . Huang, G. Xu, Q. Wang, X. Song, and X. Wang, “Efficient and privacy-preserving authentication for federated learning in industrial internet of things data sharing application,”IEEE Internet of Things Journal, vol. 12, no. 9, pp. 11 652–11 663, 2024

    2024

  22. [22]

    Enhancing Byzantine robustness of federated learning via tripartite adaptive authentication,

    X. Li, Y . Li, H. Wan, and C. Wang, “Enhancing Byzantine robustness of federated learning via tripartite adaptive authentication,”Journal of Big Data, vol. 12, no. 1, p. 121, 2025

    2025

  23. [23]

    IoT authentication in federated learning: Methods, challenges, and future directions,

    A. Badhib, S. Alshehri, and A. Cherif, “IoT authentication in federated learning: Methods, challenges, and future directions,”Sensors, vol. 25, no. 24, p. 7619, 2025

    2025

  24. [24]

    A secure drone-to-drone communication and software defined drone network-enabled traffic monitoring system,

    A. Kumar, A. S. Yadav, S. S. Gill, H. Pervaiz, Q. Ni, and R. Buyya, “A secure drone-to-drone communication and software defined drone network-enabled traffic monitoring system,”Simulation Modelling Prac- tice and Theory, 2022

    2022

  25. [25]

    Blockchain- assisted secure and lightweight authentication scheme for multi-server internet of drones environments,

    S. Ju, H. Park, S. Son, H. Kim, Y . Park, and Y . Park, “Blockchain- assisted secure and lightweight authentication scheme for multi-server internet of drones environments,”Mathematics, vol. 12, no. 24, p. 3965, 2024

    2024

  26. [26]

    Demo: A practical testbed for decentralized federated learning on physical edge devices,

    C. Feng, N. Huber, A. Huertas Celdran, G. Bovet, and B. Stiller, “Demo: A practical testbed for decentralized federated learning on physical edge devices,” 2025. [Online]. Available: https://arxiv.org/abs/2505.08033

    Pith/arXiv arXiv 2025

  27. [27]

    Federated dropout—a simple ap- proach for enabling federated learning on resource-constrained devices,

    D. Wen, K.-J. Jeon, and K. Huang, “Federated dropout—a simple ap- proach for enabling federated learning on resource-constrained devices,” IEEE Wireless Communications Letters, vol. 11, no. 5, pp. 923–927, 2022

    2022

  28. [28]

    Stars-fl: Accelerating federated learning over heterogeneous mobile devices via spatial-temporal aware reconfigured subnetworks

    B. Li, H.-a. Su, L. Li, J. Geng, X. Qin, Y . Hou, X. Fu, and M. Pan, “Stars-fl: Accelerating federated learning over heterogeneous mobile devices via spatial-temporal aware reconfigured subnetworks.”

  29. [29]

    Federated learning at the network edge: When not all nodes are created equal,

    F. Malandrino and C. F. Chiasserini, “Federated learning at the network edge: When not all nodes are created equal,”IEEE Communications Magazine, vol. 59, no. 11, pp. 80–85, 2021

    2021

  30. [30]

    Deep federated learning: A systematic review of methods, applications, and challenges,

    L. Cooray, J. Sendanayake, P. Vithanaarachchi, and Y . H. P. P. Priyadar- shana, “Deep federated learning: A systematic review of methods, applications, and challenges,”Frontiers in Computer Science, vol. 7, 2025

    2025

  31. [31]

    Federated learning in edge computing: Advancements, security challenges, and optimization strategies,

    H. Janardhanan, “Federated learning in edge computing: Advancements, security challenges, and optimization strategies,” inIEEE International Conference on Circuits, Power and Computing Technologies (ICCPCT), 2025

    2025

  32. [32]

    Federated learning for a dynamic edge: A modular and resilient approach,

    L. Almeida, R. Teixeira, G. Baldoni, M. Antunes, and R. L. Aguiar, “Federated learning for a dynamic edge: A modular and resilient approach,”Sensors, vol. 25, no. 12, p. 3812, 2025

    2025

  33. [33]

    Benchmarking edge AI platforms: Performance analysis of NVIDIA jetson and raspberry pi 5 with coral TPU,

    D. Minott, S. Siddiqui, and R. J. Haddad, “Benchmarking edge AI platforms: Performance analysis of NVIDIA jetson and raspberry pi 5 with coral TPU,” inProc. IEEE SoutheastCon, 2025, pp. 1384–1389

    2025

  34. [34]

    An experimental analysis of the WPA3 protocol in IoT devices,

    G. A. Alghisi and F. Gringoli, “An experimental analysis of the WPA3 protocol in IoT devices,” in2024 22nd Mediterranean Communication and Computer Networking Conference (MedComNet). IEEE, 2024, pp. 1–4

    2024