Pith. sign in

REVIEW 4 major objections 4 minor 45 references

This paper argues that intent-based networking for industrial IoT can validate network strategies before deployment through a strategy-similarity-aware federated mechanism, and that this mechanism improves accuracy, accelerates convergence,

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

2026-08-03 19:36 UTC pith:IYZRSIZC

load-bearing objection A credible combination of known ideas that would be worth building on — but the empirical support rests on an undefined label construction, so read the headline numbers as provisional. the 4 major comments →

arxiv 2512.20627 v2 pith:IYZRSIZC submitted 2025-11-28 cs.NI cs.AI

Efficient Asynchronous Federated Evaluation with Strategy Similarity Awareness for Intent-Based Networking in Industrial Internet of Things

Shaowen Qin , Jianfeng Zeng , Haodong Guo , Xiaohuan Li , Jiawen Kang , Qian Chen This is my paper
classification cs.NI cs.AI
keywords intent-based networkingfederated learningindustrial internet of thingsstrategy similarityasynchronous aggregationcommunication efficiencymultimodal intent alignmentpolicy verification
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 proposes that strategy verification for intent-based networking in industrial IoT can be done as a federated regression task, where each node privately judges whether a candidate strategy suits its local operation. Its main claim is that a strategy-similarity-aware selection mechanism plus thresholded asynchronous uploads (SSAFL) improves model accuracy and convergence speed while reducing communication cost by 27.8% compared with SemiAsyn. A sympathetic reader would care because this promises pre-deployment validation of intents without raw data leaving nodes or incurring costly rollbacks.

Core claim

FEIBN is an end-to-end framework that uses pretrained encoders and a large language model to convert multimodal user intents (text, voice, images) into structured strategy tuples, then runs a federated regression task that outputs a deployability score for a candidate strategy before it is deployed. Within this framework, SSAFL computes a strategy similarity score for each node as a weighted combination of action-set overlap (Jaccard) and condition-threshold closeness (exponential decay), combines it with a resource availability score into a suitability score, and uses a threshold on that score to select participating nodes. Uploads are asynchronous and only occur when the local model update

What carries the argument

The strategy similarity metric Sim_i(S) is the central object: it scores semantic closeness between the candidate strategy and a node's historical strategies via Jaccard similarity on action sets and an exponential-decay pairwise condition similarity on goal thresholds. The suitability score H_i couples this similarity with a resource score built from CPU and bandwidth, driving node selection; the personalized upload threshold ε_i makes high-similarity nodes upload more eagerly. These two mechanisms—similarity-aware selection and thresholded asynchronous upload—carry the argument for communication savings without convergence loss, supported by an asymptotic convergence proof under smoothness

Load-bearing premise

Each node's local dataset contains a trustworthy label indicating whether a candidate strategy is suitable in that node's context, and those labels reflect what would actually happen if the strategy were deployed; the paper does not describe how these labels are produced or verified on a real deployment.

What would settle it

If a testbed experiment replaced the paper's unspecified suitability labels with measured deployment outcomes (e.g., the empirical satisfaction probability p_S from Eq. (10)) and reran SSAFL against SemiAsyn, and SSAFL no longer cut communication cost by roughly 27.8% or matched its reported R² gains, the central claim would be falsified. More directly, any deployment where thresholded asynchronous uploads with similarity weighting fails to improve convergence over plain asynchronous FL would contradict the claimed mechanism.

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

If this is right

  • If SSAFL works as claimed, IIoT operators can validate strategies pre-deployment over federated data, cutting downtime risk from faulty policy rollouts.
  • The 27.8% communication saving scales directly to bandwidth-constrained industrial networks, making federated evaluation practical at larger node counts.
  • Strategy-similarity prioritization generalizes to other FL tasks where clients have heterogeneous historical relevance, not just IBN.
  • The personalized threshold design offers a tunable trade-off between communication reduction and model quality that operators can set per scenario.
  • The convergence analysis in the appendix indicates the similarity-aware thresholding keeps standard asynchronous FL convergence rates when trigger bias stays small.

Where Pith is reading between the lines

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

  • The paper does not describe how the local suitability labels y_j are constructed; if those labels come from simulated or proxy criteria rather than measured deployment outcomes, the reported gains could shrink on real factory floors.
  • The strategy tuple translation by an LLM is upstream of the similarity metric; translation errors would propagate into node selection, so an end-to-end test should couple translation quality with SSAFL performance.
  • In a real IIoT loop, communication cost is only part of the overhead; the event-window aggregation and threshold tuning add server-side complexity that the paper does not fully cost.
  • The same selection-plus-threshold mechanism could be applied to safety-critical configuration validation in other domains, provided a trustworthy local suitability signal exists.

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

4 major / 4 minor

Summary. The paper proposes FEIBN, an intent-based networking framework for IIoT that uses LLMs to translate multimodal user intents into structured strategy tuples and federated learning to evaluate those strategies across distributed nodes without exposing local data. The key algorithmic contribution is SSAFL, a strategy-similarity-aware asynchronous federated learning mechanism that selects nodes by combining a strategy similarity score (action, condition, resource components) with a resource availability score, and applies thresholded asynchronous uploads to reduce communication cost. The paper reports experiments on a regression task built from Edge-IIoTset and 'intent-related data', comparing SSAFL with FedAvg, FedAsyn, and SemiAsyn. The central claim is that SSAFL improves model accuracy, accelerates convergence, and reduces communication cost by 27.8% compared with SemiAsyn. The paper also provides a convergence analysis in Appendix A under smoothness, bounded staleness, and a trigger-bias control assumption (A6).

Significance. If the claimed results are credible, the paper would make a practical contribution by coupling LLM-based intent translation with privacy-preserving federated strategy verification, and the SSAFL selection/upload mechanism could be useful for communication-constrained IIoT. The paper also presents an explicit convergence framework with theorems in the convex and PL settings, which is a positive feature. However, the credibility of the empirical claims rests on an unspecified label construction for the regression task, and the headline 27.8% communication-cost reduction is not supported by any reported aggregate number. The convergence proof depends on an unproved Assumption A6 that is itself load-bearing. These gaps prevent the current manuscript from providing a solid basis for the central claims.

major comments (4)
  1. [§III-C, Eq. (4), §V.A] The experimental proxy is not specified sufficiently. The local dataset D_i is defined with labels y_j 'indicating whether policy S is suitable under the local context', but the paper never describes how y_j is constructed from Edge-IIoTset or from the 'intent-related data' mentioned in §V.A. If y_j is generated by a deterministic rule or a heuristic score, the regression task may be trivially learnable and the accuracy/convergence differences in Table IV may not reflect real IBN policy-verification quality. Please provide an explicit label-generation procedure, or use real deployment outcome data with a disclosed provenance.
  2. [Abstract, §V.C, Fig. 7] The abstract claims 'reduce the cost by 27.8% compared with SemiAsyn', but no table or figure reports an aggregate communication cost or percentage reduction. Fig. 7 shows communication rounds for only three clients and gives no overall percentage. Since this is a headline quantitative claim, report the total communication cost (e.g., sum of client rounds or transmitted bytes) for each method or provide a table with the required percentage and variance.
  3. [Appendix A, Assumption A6, Lemma 1, Theorems 1–2] The convergence analysis relies on Assumption A6, which asserts that the deviation between pre-weight and post-weight updates is bounded by ζ times the pre-weight update norm. This is asserted as 'mild' without proof. The min-weight protection and renormalization in Eqs. (18)–(19) can change the aggregation direction significantly, especially when several small-magnitude updates are boosted to w_min and then renormalized, so the O(ζ) terms in Lemma 1 and Theorems 1–2 may be large. Please provide a rigorous derivation of A6 from the algorithm, or replace it with a proved bound under explicit conditions on w_min, ϵ_base, and λ_s.
  4. [Table IV, §V.C] The accuracy improvement over SemiAsyn is not statistically supported by the reported error bars. SSAFL reports MAE 0.0497±0.011 and R² 0.9177±0.12 versus SemiAsyn's MAE 0.0594±0.017 and R² 0.8840±0.11. With n=5, these standard deviations overlap substantially, and no significance test is provided. The claim 'improves model accuracy' needs confidence intervals, paired tests, or a clear statement of the effect size's uncertainty.
minor comments (4)
  1. [§IV.A, Eq. (11)] The formatting of Eq. (11) is broken in the manuscript; also the definition of condition similarity uses both c and g inconsistently. Please rewrite the equation and reconcile the notations (C, c, g, G) to improve readability.
  2. [§V.C, Fig. 6 and Table IV] The text states 'FedAvg and FedAsyn show slower convergence and lower final R² scores, around 0.85 and 0.83 respectively', but Table IV lists FedAvg R²=0.8398 and FedAsyn R²=0.7462. These numbers are inconsistent; please correct one source.
  3. [Eqs. (17)–(18)] The aggregation weight notation is confusing: w'_i is defined as a product but then appears in a max and a normalization with w_min. Clarify the units and whether w'_i is pre-normalization or a raw score.
  4. [General] The paper does not mention any plan to release code or data. Given the dependence of the results on private/synthetic label construction, a release or a detailed supplemental appendix would materially aid reproducibility.

Circularity Check

0 steps flagged

No load-bearing circularity; SSAFL's gains are validated against external baselines, and self-citations are confined to related work.

full rationale

The paper's central claim—SSAFL improves model accuracy, accelerates convergence, and reduces communication cost—is supported by comparisons against external baselines (FedAvg, FedAsyn, SemiAsyn) on Edge-IIoTset plus the authors' intent-related data, rather than by fitting parameters to the target result. The selection score H_i (Eq. 13), upload threshold epsilon_i (Eq. 16), and aggregation weight w'_i (Eq. 17) all use the same strategy-similarity measure Sim_i(S); this is a deliberate algorithmic design, not a circular derivation, because the regression target y_j in Eq. (4) is not defined as a function of Sim_i(S). The convergence analysis in Appendix A is conditional on standard assumptions plus Assumption A6, which postulates that thresholded uploads and renormalization introduce only a bounded bias zeta, and then carries O(zeta) into the final bounds. A6 is asserted rather than derived, so the proof is not fully self-contained; however, this is an omitted-support/proof-gap concern, not a reduction of the theorem to its own conclusion. Similarly, the construction of the suitability labels y_j is not specified, leaving the external validity of the experiments open to question; that too is a correctness risk rather than circularity. The only self-citations (e.g., [7] and [23]) appear in the introduction and related-work discussion and are not load-bearing for the SSAFL results or the convergence analysis. No uniqueness theorem or ansatz is imported from the authors' prior work, and no reported prediction is equivalent by construction to an input. Therefore no specific circular step can be quoted.

Axiom & Free-Parameter Ledger

9 free parameters · 7 axioms · 2 invented entities

SSAFL rests on standard smoothness and staleness assumptions, plus a domain proxy: that regression labels in the private dataset measure real strategy suitability. It also assumes A6, a bound on thresholding bias that is asserted rather than derived. The design introduces several hand-set weights and thresholds (γ, β, δ, τ_s, ϵ_base, λ_s, w_min, α_g, event window) with no sensitivity analysis.

free parameters (9)
  • γ1, γ2
    Weights for action versus condition similarity in Eq. (11), chosen by hand with γ1+γ2=1; no sensitivity analysis is reported.
  • β1, β2
    Weights for strategy similarity versus resource availability in Eq. (13), chosen by hand with β1+β2=1.
  • δ1, δ2
    Weights for CPU availability and bandwidth in Eq. (14), chosen by hand with δ1+δ2=1.
  • τ_s
    Suitability-score threshold for selecting nodes in Section IV-A; the value is not reported and no sensitivity study is given.
  • ϵ_base
    Base upload threshold in Eq. (16); hand-chosen and not reported.
  • λ_s
    Scaling factor controlling similarity's effect on the upload threshold in Eq. (16); hand-chosen.
  • w_min
    Minimum aggregation weight in Eq. (18); hand-chosen.
  • α_g (a_g)
    Scaling factor in the exponential condition-similarity function h(g,g′) in Eq. (12); hand-chosen.
  • Δ (event window)
    Micro-batch aggregation window in Algorithm 2; hand-chosen and not reported.
axioms (7)
  • standard math A1/A2: Local objectives are L-smooth and local stochastic gradients are unbiased with bounded variance.
    Standard assumptions in FL convergence analysis, stated in Appendix A.
  • domain assumption A3: Bounded staleness τ_max in asynchronous updates.
    Assumes network delays keep gradient staleness bounded; not verified in the experiments.
  • domain assumption A4: Each client uses a constant step size and finite local steps.
    Algorithmic choice assumed in the convergence proof, Appendix A.
  • domain assumption A5: Aggregation weights satisfy w_i ≥ w_min > 0 and sum to 1.
    Enforced by minimum-weight protection and renormalization in Algorithm 2.
  • ad hoc to paper A6: Trigger bias control — the deviation between pre-weight and post-weight updates is bounded by ζ < 1.
    Assumes the very property of the thresholding mechanism that the paper needs for convergence; no explicit bound in terms of ϵ_base and λ_s is proven.
  • domain assumption Each local dataset D_i contains valid suitability labels y_j for whether a strategy is appropriate in the node's context.
    Invoked in Sections III-C and V.A; the label construction is not described and no real deployment validates it.
  • domain assumption Edge-IIoTset plus the private intent-related data adequately represents IBN strategy verification in IIoT.
    Section V.A states this representativeness without empirical support; the paper asserts the same qualitative trends would hold on other datasets.
invented entities (2)
  • Strategy similarity score Sim_i(S) no independent evidence
    purpose: Used in node selection, upload thresholds, and aggregation weighting to quantify relevance of a node's historical strategies to the current strategy.
    Defined by Eqs. (11)-(12) with hand-chosen weights; no external benchmark or real deployment data validates that it truly captures policy relevance.
  • Suitability/participation score H_i no independent evidence
    purpose: Combines strategy similarity and resource availability to select participating nodes.
    Defined by Eq. (13); not validated independently of SSAFL's own pipeline.

pith-pipeline@v1.3.0-alltime-deepseek · 17927 in / 15588 out tokens · 160530 ms · 2026-08-03T19:36:47.376502+00:00 · methodology

0 comments
read the original abstract

Intent-Based Networking (IBN) offers a promising paradigm for intelligent and automated network control in Industrial Internet of Things (IIoT) environments by translating high-level user intents into executable network strategies. However, frequent strategy deployment and rollback are impractical due to tightly coupled workflows and high downtime costs, while node heterogeneity and privacy constraints further complicate centralized strategy evaluation. To address these challenges, we propose a Federated Evaluation Enhanced Intent-Based Networking framework (FEIBN), which leverages large language models (LLMs) to translate user intents into structured strategy tuples and employs federated learning to support distributed strategy evaluation. To improve training efficiency and reduce communication overhead, we design a Strategy Similarity Aware Federated Learning mechanism (SSAFL), which selects nodes relevant to the task based on strategy similarity and resource status, and triggers asynchronous model uploads only when local updates are significant. Experiments demonstrate that the proposed method improves model accuracy, accelerates convergence, and reduces communication cost compared with the baselines.

Figures

Figures reproduced from arXiv: 2512.20627 by Haodong Guo, Jianfeng Zeng, Jiawen Kang, Qian Chen, Shaowen Qin, Xiaohuan Li.

Figure 1
Figure 1. Figure 1: The FEIBN framework for IIoT. The framework supports intent-driven strategy generation and verification across distributed IIoT nodes. Within [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The process of SSAFL. Nodes first evaluate their strategy similarity [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Matching accuracy of strategy tuples under different LLMs. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Federated evaluation number under different matching accuracies. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Model training curves of different methods. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Communication rounds of clients under different methods. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 5
Figure 5. Figure 5: shows the total time required for strategy deploy￾ment across different methods. Adding only the alignment module slightly increases the deployment time due to the additional semantic parsing process. In contrast, FEIBN that integrates both alignment and federated evaluation results in a higher overall time cost, especially under lower alignment accuracy such as FEIBN-0.6, where more verification rounds ar… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

45 extracted references · 4 linked inside Pith

  1. [1]

    Industrial internet of things intelligence empowering smart manufacturing: A literature review,

    Y . Hu, Q. Jia, Y . Yao, Y . Lee, M. Lee, C. Wang, X. Zhou, R. Xie, and F. R. Yu, “Industrial internet of things intelligence empowering smart manufacturing: A literature review,”IEEE Internet of Things Journal, vol. 11, no. 11, pp. 19 143–19 167, 2024

    2024

  2. [2]

    Industrial internet of things: Requirements, architecture, challenges, and future research directions,

    M. Alabadi, A. Habbal, and X. Wei, “Industrial internet of things: Requirements, architecture, challenges, and future research directions,” IEEE Access, vol. 10, pp. 66 374–66 400, 2022

    2022

  3. [3]

    Intent-based networking: Current advances, open challenges, and future directions,

    M. Gharbaoui, B. Martini, and P. Castoldi, “Intent-based networking: Current advances, open challenges, and future directions,” in2023 23rd International Conference on Transparent Optical Networks (ICTON), 2023, pp. 1–5

    2023

  4. [4]

    A survey on intent-based networking,

    A. Leivadeas and M. Falkner, “A survey on intent-based networking,” IEEE Communications Surveys & Tutorials, vol. 25, no. 1, pp. 625–655, 2023

    2023

  5. [5]

    Intent-based approaches for industry 4.0 applications: A systematic mapping study,

    K. Sadouki and E. Kornyshova, “Intent-based approaches for industry 4.0 applications: A systematic mapping study,”Internet of Things, vol. 32, p. 101629, 2025. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S254266052500143X

    2025

  6. [7]

    Full-life cycle intent-driven network verification: Challenges and approaches,

    Y . Song, C. Yang, J. Zhang, X. Mi, and D. Niyato, “Full-life cycle intent-driven network verification: Challenges and approaches,”IEEE Network, vol. 37, no. 5, pp. 145–153, 2023

    2023

  7. [8]

    Network copilot: Intent-driven network configuration updating for service guarantee,

    R. Han, J. Wang, H. Sun, Z. Jiang, Q. Qi, Z. Zhuang, Y . Zhang, and J. Liao, “Network copilot: Intent-driven network configuration updating for service guarantee,” inIEEE INFOCOM 2025-IEEE Conference on Computer Communications. IEEE, 2025, pp. 1–10

    2025

  8. [9]

    In-context retrieval-augmented language mod- els,

    O. Ram, Y . Levine, I. Dalmedigos, D. Muhlgay, A. Shashua, K. Leyton- Brown, and Y . Shoham, “In-context retrieval-augmented language mod- els,”Transactions of the Association for Computational Linguistics, vol. 11, pp. 1316–1331, 2023

    2023

  9. [10]

    Following the compass: Llm-empowered intent translation with manual guidance,

    L. Guo, J. Wang, J. Wu, C. Yan, H. Sun, Z. Zhuang, Q. Qi, Y . Dong, H. Ren, and J. Liao, “Following the compass: Llm-empowered intent translation with manual guidance,” in2024 IEEE 32nd International Conference on Network Protocols (ICNP), 2024, pp. 1–12

    2024

  10. [11]

    Human-centred ai in industry 5.0: a systematic review,

    M. Passalacqua, R. Pellerin, F. Magnani, P. Doyon-Poulin, L. Del- Aguila, J. Boasen, and P.-M. L ´eger, “Human-centred ai in industry 5.0: a systematic review,”International Journal of Production Research, vol. 63, no. 7, pp. 2638–2669, 2025

    2025

  11. [12]

    Intent-driven secure system design: Methodology and implementation,

    S. E. Ooi, R. Beuran, T. Kuroda, T. Kuwahara, R. Hotchi, N. Fujita, and Y . Tan, “Intent-driven secure system design: Methodology and implementation,”Computers & Security, vol. 124, p. 102955, 2023

    2023

  12. [13]

    A survey on security and cryptographic perspective of industrial-internet-of-things,

    N. Mishra, S. H. Islam, and S. Zeadally, “A survey on security and cryptographic perspective of industrial-internet-of-things,”Internet of Things, vol. 25, p. 101037, 2024

    2024

  13. [14]

    Federated learning: Strategies for improving communication efficiency,

    J. Kone ˇcn`y, H. B. McMahan, F. X. Yu, P. Richt ´arik, A. T. Suresh, and D. Bacon, “Federated learning: Strategies for improving communication efficiency,”arXiv preprint arXiv:1610.05492, 2016

    Pith/arXiv arXiv 2016

  14. [15]

    Communication-efficient learning of deep networks from decentralized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” inArtificial intelligence and statistics. PMLR, 2017, pp. 1273– 1282

    2017

  15. [16]

    Decentralized federated learning: A survey and perspective,

    L. Yuan, Z. Wang, L. Sun, P. S. Yu, and C. G. Brinton, “Decentralized federated learning: A survey and perspective,”IEEE Internet of Things Journal, vol. 11, no. 21, pp. 34 617–34 638, 2024

    2024

  16. [17]

    Convergence analysis of sequential federated learning on heterogeneous data,

    Y . Li and X. Lyu, “Convergence analysis of sequential federated learning on heterogeneous data,”Advances in Neural Information Processing Systems, vol. 36, pp. 56 700–56 755, 2023

    2023

  17. [18]

    Every parameter matters: Ensuring the convergence of federated learning with dynamic heterogeneous models reduction,

    H. Zhou, T. Lan, G. P. Venkataramani, and W. Ding, “Every parameter matters: Ensuring the convergence of federated learning with dynamic heterogeneous models reduction,”Advances in Neural Information Pro- cessing Systems, vol. 36, pp. 25 991–26 002, 2023

    2023

  18. [19]

    Fedasmu: Efficient asynchronous federated learning with dynamic staleness-aware model update,

    J. Liu, J. Jia, T. Che, C. Huo, J. Ren, Y . Zhou, H. Dai, and D. Dou, “Fedasmu: Efficient asynchronous federated learning with dynamic staleness-aware model update,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 12, 2024, pp. 13 900–13 908

    2024

  19. [20]

    Enhancing network management using code generated by large language models,

    S. K. Mani, Y . Zhou, K. Hsieh, S. Segarra, T. Eberl, E. Azulai, I. Frizler, R. Chandra, and S. Kandula, “Enhancing network management using code generated by large language models,” inProceedings of the 22nd ACM Workshop on Hot Topics in Networks, 2023, pp. 196–204

    2023

  20. [21]

    Prosper: Extracting protocol specifica- tions using large language models,

    P. Sharma and V . Yegneswaran, “Prosper: Extracting protocol specifica- tions using large language models,” inProceedings of the 22nd ACM Workshop on Hot Topics in Networks, 2023, pp. 41–47

    2023

  21. [22]

    Intent-based networking and its application to optical networks,

    L. Velasco, S. Barzegar, F. Tabatabaeimehr, and M. Ruiz, “Intent-based networking and its application to optical networks,”Journal of optical communications and networking, vol. 14, no. 1, pp. A11–A22, 2021

    2021

  22. [23]

    An adaptive asynchronous federated learning framework for heterogeneous internet of things,

    W. Zhang, D. Deng, X. Wu, W. Zhao, Z. Liu, T. Zhang, J. Kang, and D. Niyato, “An adaptive asynchronous federated learning framework for heterogeneous internet of things,”Information Sciences, vol. 689, p. 121458, 2025

    2025

  23. [24]

    An ai-driven intent-based network architecture,

    Y . Njah, A. Leivadeas, and M. Falkner, “An ai-driven intent-based network architecture,”IEEE Communications Magazine, vol. 63, no. 4, pp. 146–153, 2025

    2025

  24. [25]

    Towards intent-based network management: Large language models for intent extraction in 5g core networks,

    D. M. Manias, A. Chouman, and A. Shami, “Towards intent-based network management: Large language models for intent extraction in 5g core networks,” in2024 20th International Conference on the Design of Reliable Communication Networks (DRCN), 2024, pp. 1–6

    2024

  25. [26]

    Intent assurance using llms guided by intent drift,

    K. Dzeparoska, A. Tizghadam, and A. Leon-Garcia, “Intent assurance using llms guided by intent drift,” inNOMS 2024-2024 IEEE Network Operations and Management Symposium, 2024, pp. 1–7

    2024

  26. [27]

    Agentic ai for intent-based industrial automation,

    M. Lima Romero and R. Suyama, “Agentic ai for intent-based industrial automation,”arXiv e-prints, pp. arXiv–2506, 2025

    2025

  27. [28]

    Confidence-based similarity- aware personalized federated learning for autonomous iot,

    X. Han, Q. Zhang, Z. He, and Z. Cai, “Confidence-based similarity- aware personalized federated learning for autonomous iot,”IEEE Inter- net of Things Journal, vol. 11, no. 7, pp. 13 070–13 081, 2023

    2023

  28. [29]

    Loci: Federated continual learning of heterogeneous tasks at edge,

    Y . Luopan, R. Han, Q. Zhang, X. Zuo, C. H. Liu, G. Wang, and L. Y . Chen, “Loci: Federated continual learning of heterogeneous tasks at edge,”IEEE Transactions on Parallel and Distributed Systems, 2025

    2025

  29. [30]

    Pfedsa: Personalized federated multi-task learning via similarity awareness,

    C. Ye, H. Zheng, Z. Hu, and M. Zheng, “Pfedsa: Personalized federated multi-task learning via similarity awareness,” in2023 IEEE International Parallel and Distributed Processing Symposium (IPDPS). IEEE, 2023, pp. 480–488

    2023

  30. [31]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” inPro- ceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), 2019, pp. 4171–4186

    2019

  31. [32]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in neural information processing systems, vol. 33, pp. 12 449– 12 460, 2020

    2020

  32. [33]

    Deep residual neural-network-based robot joint fault diagnosis method,

    J. Pan, L. Qu, and K. Peng, “Deep residual neural-network-based robot joint fault diagnosis method,”Scientific Reports, vol. 12, no. 1, p. 17158, 2022

    2022

  33. [34]

    Gpt-5 and open-weight large language models: Advances in reasoning, transparency, and control,

    M. Leon, “Gpt-5 and open-weight large language models: Advances in reasoning, transparency, and control,”Information Systems, p. 102620, 2025

    2025

  34. [35]

    Deepseek-v3 technical report,

    A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruanet al., “Deepseek-v3 technical report,”arXiv preprint arXiv:2412.19437, 2024

    Pith/arXiv arXiv 2024

  35. [36]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosaleet al., “Llama 2: Open foundation and fine-tuned chat models,”arXiv preprint arXiv:2307.09288, 2023. 13

    Pith/arXiv arXiv 2023

  36. [37]

    A survey on kernel-based multi-task learning,

    C. Ruiz, C. M. Ala ´ız, and J. R. Dorronsoro, “A survey on kernel-based multi-task learning,”Neurocomputing, vol. 577, p. 127255, 2024

    2024

  37. [38]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,”Proceedings of Machine learning and systems, vol. 2, pp. 429–450, 2020

    2020

  38. [39]

    Federated machine learning: Concept and applications,

    Q. Yang, Y . Liu, T. Chen, and Y . Tong, “Federated machine learning: Concept and applications,”ACM Transactions on Intelligent Systems and Technology (TIST), vol. 10, no. 2, pp. 1–19, 2019

    2019

  39. [40]

    Asynchronous federated optimization,

    C. Xie, S. Koyejo, and I. Gupta, “Asynchronous federated optimization,” arXiv preprint arXiv:1903.03934, 2019

    Pith/arXiv arXiv 1903

  40. [41]

    Edge-iiotset: A new comprehensive realistic cyber security dataset of iot and iiot applications for centralized and federated learning,

    M. A. Ferrag, O. Friha, D. Hamouda, L. Maglaras, and H. Janicke, “Edge-iiotset: A new comprehensive realistic cyber security dataset of iot and iiot applications for centralized and federated learning,”IEEE Access, vol. 10, pp. 40 281–40 306, 2022

    2022

  41. [42]

    Communication-Efficient Learning of Deep Networks from Decentralized Data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y. Arcas, “Communication-Efficient Learning of Deep Networks from Decentralized Data,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics, ser. Proceedings of Machine Learning Research, A. Singh and J. Zhu, Eds., vol. 54. PMLR, 20–22 Apr 2017, pp. 1273–1282. [...

    2017

  42. [43]

    Safa: A semi- asynchronous protocol for fast federated learning with low overhead,

    W. Wu, L. He, W. Lin, R. Mao, C. Maple, and S. Jarvis, “Safa: A semi- asynchronous protocol for fast federated learning with low overhead,” IEEE Transactions on Computers, vol. 70, no. 5, pp. 655–668, 2021

    2021

  43. [44]

    A review of federated learning methods in heterogeneous scenarios,

    J. Pei, W. Liu, J. Li, L. Wang, and C. Liu, “A review of federated learning methods in heterogeneous scenarios,”IEEE Transactions on Consumer Electronics, vol. 70, no. 3, pp. 5983–5999, 2024

    2024

  44. [45]

    Scheduling and aggregation design for asynchronous federated learning over wireless networks,

    C.-H. Hu, Z. Chen, and E. G. Larsson, “Scheduling and aggregation design for asynchronous federated learning over wireless networks,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 4, pp. 874–886, 2023

    2023

  45. [46]

    Fedsa: A staleness-aware asynchronous federated learning algorithm with non-iid data,

    M. Chen, B. Mao, and T. Ma, “Fedsa: A staleness-aware asynchronous federated learning algorithm with non-iid data,”Future Generation Computer Systems, vol. 120, pp. 1–12, 2021

    2021