Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Edge AI can complement cloud computing to deliver low-latency, privacy-preserving, resilient civil-aviation intelligence.

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 12:00 UTC pith:3TL4Q6Z7

load-bearing objection Useful aviation-specific edge-AI survey with a solid paradigm-to-deployment mapping, but benefit claims and the certification premise need qualification. the 3 major comments →

arxiv 2607.19676 v1 pith:3TL4Q6Z7 submitted 2026-07-22 cs.AI

Edge Intelligence in Civil Aviation: Paradigms, Techniques, and Applications

Wenbin Li , Zhongtian Liao , Bolin Liu , Yongjie Zhou , Jingling Wu , Xiaoyong Lin , Jing Chen This is my paper
classification cs.AI
keywords edge intelligencecivil aviationmodel compressionfederated learningsplit learningcloud-edge collaborationedge inferencesafety-critical AI
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.

This paper argues that civil aviation's growing reliance on cloud-hosted AI is mismatched with the industry's safety-critical, latency-sensitive, and connectivity-limited operations. It claims that moving perception, prediction, and decision logic toward the network edge — onboard aircraft, at towers, and on airport premises — can sharply cut response time, keep sensitive data local, and keep AI working during communication outages. The paper systematizes the enabling techniques (quantization, pruning, distillation, federated and split learning) and four computing patterns for distributing AI across devices, edges, and cloud. A sympathetic reader would take away that edge intelligence is not a replacement for cloud but a complementary layer that makes aviation AI faster, more private, and more resilient.

Core claim

The paper argues that civil aviation generates large volumes of safety-critical data at the network edge — flight decks, towers, ramps, maintenance bays — and that cloud-centric AI, which moves that data to remote data centers for inference, is a poor fit for operations that demand low latency, can lose connectivity, and treat data privacy as a regulatory requirement. It claims that edge intelligence, built from model compression (quantization, pruning, distillation), collaborative inference, and distributed learning (federated and split learning), can push perception and decision logic closer to the data source. The paper's central thesis: a refined edge solution can complement cloud founda

What carries the argument

The load-bearing organizing device is a taxonomy of four computing patterns for distributing AI across devices, edge servers, and cloud: cloud-enabled (device collaborates with a distant cloud), edge-enabled (device collaborates with a local edge server, no cloud), hierarchical (device-edge-cloud all participate), and peer-to-peer (devices collaborate directly without a server). The paper pairs each pattern with a technique stack — quantization, pruning, knowledge distillation, early exit, split computing, federated learning, split learning — that makes models small and efficient enough to run on resource-limited devices. The taxonomy maps each pattern to concrete aviation deployment loci (a

Load-bearing premise

The argument assumes that compressed, federated, or split models can run on certified avionics-grade hardware and airport edge servers with accuracy and reliability high enough for safety-critical tasks, yet the paper offers no certification evidence or measured performance data.

What would settle it

A field trial on certified avionics hardware: run a 4-bit quantized runway-incursion detection model at a busy airport during live operations and record end-to-end latency and false-alarm rate. If latency or error rates exceed the operational safety budget that a cloud-based system would meet, the paper's central claim that edge models can complement or replace cloud inference in safety-critical loops is falsified.

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

If this is right

  • Tower assistance and runway-incursion alerts can operate in the sub-100 ms range locally, without waiting for cloud round-trips.
  • Sensitive operational data (voice, video, telemetry) can stay on site or on board, reducing privacy and data-sovereignty exposure.
  • AI services keep functioning on oceanic and polar routes and at remote aerodromes during satcom or ground-link outages.
  • Fleet-level model improvement can proceed through federated or split learning without pooling raw data across airlines.
  • If edge-capable foundation models deliver on the compression claims, cockpit and tower assistants could run on-premises with cloud used only for periodic model refreshes.

Where Pith is reading between the lines

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

  • If the paper's argument holds, the certification burden in aviation may shift from isolated model accuracy to end-to-end system assurance, with uncertainty estimates and reject options becoming standard edge outputs.
  • A natural testable extension is a field comparison at a non-towered airport: measure detection latency and accuracy of a quantized vision model under an emulated backhaul outage against a cloud-only baseline.
  • The four-pattern taxonomy likely generalizes beyond aviation to other safety-critical, intermittently connected domains such as maritime navigation and rail operations, though the paper does not make this claim.

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. This manuscript is a survey/position paper arguing that edge AI—model compression, federated/split learning, and edge-cloud orchestration—should complement cloud-based AI in civil aviation. It organizes the material into edge-inference and edge-learning techniques (Section II), four organizational paradigms (cloud-enabled, edge-enabled, hierarchical, peer-to-peer) with a deployment table (Section III and Table II), and representative applications across ATC, apron surveillance, predictive maintenance, non-towered airports, and UAV/UTM (Section IV). It closes with six future trends, including edge-capable foundation models, federated/split learning at scale, network-native orchestration, certification by design, robustness to missing modalities, and synthetic-data stress testing (Section V). The central claim is that a refined edge solution can deliver low-latency, privacy-preserving, and resilient AI services across the civil-aviation lifecycle.

Significance. The taxonomy of four computing paradigms mapped to aviation deployment loci is a useful organizing contribution, and the paper brings together a broad set of recent ML-compression and edge-computing references that are not commonly combined in the aviation literature. The strengths include the explicit mapping of generic edge-AI techniques to aviation-specific settings (Tables I and II), the identification of certification as a key gap, and the proposal of concrete future research directions (Section V). However, the paper does not present new experimental results or a systematic evidence review; its main value is as a scoping document and research agenda. If its central claim is read as an established engineering fact, that claim is currently unsupported: the benefits of sub-second latency, offline resilience, and privacy preservation are asserted from generic ML results and a few academic prototypes, not from certified avionics deployments. The significance is therefore conditional on reframing the paper as a research prospectus rather than a validated deployment blueprint.

major comments (3)
  1. [Section II-C (Fig. 1) and Introduction] The paper asserts that edge AI 'achieves sub-second reactions' and cites the '$600 per minute of delay' figure from a trade-press article [8]. No primary source or peer-reviewed measurement is provided for either claim. The $600 figure is a widely quoted industry anecdote and should be replaced with a citable cost model or explicitly labeled as an illustrative estimate. The 'sub-second reactions' claim is also not tied to any measured end-to-end latency on aviation hardware in the cited literature. Because these two claims are used to motivate the entire paper, they are load-bearing and must be either substantiated or softened to 'can potentially achieve' with concrete evidence.
  2. [Section V-A and Table II] The central premise—that compressed, federated, or split models can run on certified avionics-grade hardware or airport edge servers with sufficient accuracy and reliability for safety-critical tasks—is asserted without supporting evidence. Table II assigns runway/apron hazard detection and on-aircraft abnormal-event management to edge-enabled deployment, and Section V-A states that compression techniques 'make it realistic to host' such models on avionics-grade hardware. But none of the cited references (e.g., AWQ, QLoRA, SmoothQuant) provide DO-178C/DO-326A certification evidence or measured performance on aviation-qualified hardware. The paper itself acknowledges in Sections V-D and VI that certification-oriented evaluation is still future work. This is not a logical inconsistency, but it is a load-bearing unsupported assumption: the low-latency, offline, and privacy benefits all depe
  3. [Section IV-B (Applications)] The paper claims that edge intelligence is 'already delivering value' in civil aviation, citing [35]–[42]. However, these references are predominantly academic prototypes, feasibility studies, or surveys (e.g., the non-towered airport system in [42], the apron-surveillance studies in [35][36], and the UAV edge-computing surveys in [43][44]). None of them demonstrate sustained operational deployment in a certified aviation environment with quantified latency, accuracy, and reliability metrics. The language in Section IV-B ('demonstrate', 'production-grade feasibility', 'validating edge-only viability') overstates the evidence. The authors should recalibrate these claims to describe research prototypes and pilot studies, and distinguish between demonstrated deployments and promising evaluations.
minor comments (5)
  1. [Section II-C] Typo: 'this lat ency headroom' should be 'this latency headroom'. Also, the list claims 'seven core benefits' but only six bullets are enumerated (latency, privacy/security, computational efficiency, bandwidth, offline, energy, cost—actually seven; please count carefully and align the text).
  2. [References [6], [25], [35], [36]] Several references are incomplete: [6] is only an IEEE Xplore link with no author/title metadata beyond the title; [25] links to PMLR without author names; [35] and [36] have no author names or journal volume/page numbers. Provide full bibliographic details for all references.
  3. [Table II] The 'Operational latency target' for Cloud-Enabled is '100 ms–seconds', which is a wide and vague range. Also, the text in Section II-C claims 'sub-second reactions' as a general benefit, which conflicts with the cloud-enabled row's admission that it is 'not hard-real-time'. Consider aligning the terminology and providing more precise, source-backed latency ranges.
  4. [Section III and Fig. 2] The text references 'fig. 2' but Table I appears before the figure and is not explicitly referenced in the text. Add explicit cross-references (e.g., 'as summarized in Table I and illustrated in Fig. 2') to improve readability.
  5. [Section II-A] The description of Phi-3-mini includes performance figures (69% MMLU, 8.38 MT-bench, >12 tokens/s on iPhone 14) that are not aviation-specific. While relevant to on-device LLM capability, the connection to civil-aviation hardware constraints (e.g., DO-254/DO-178C qualified processors) is not made. Add a sentence clarifying the gap between consumer-device deployment and avionics-qualified deployment.

Circularity Check

0 steps flagged

No circular derivation; survey claim rests on external literature, with only a minor non-load-bearing self-citation.

full rationale

This is a survey/position paper rather than a derivation: it contains no equations, fitted parameters, or predicted quantities that reduce to its inputs. The central claim—that edge AI can complement cloud foundations for low-latency, privacy-preserving, resilient aviation services—is argued from external literature on quantization, pruning, distillation, federated/split learning, and deployed airport/UAV systems. The paper explicitly labels certification evidence as future work (§V-D, §VI), which is an evidence gap rather than a circular step. The only self-citation is [2], used for a general statement that AI can mine aviation data; the four organizational paradigms are not justified solely by [2], and the central claim does not depend on that taxonomy. Assertions such as 'models can execute on avionics-grade hardware or airport edge servers' (§V-A) are feasibility assumptions, not circular reductions. Overall, no significant circularity; the score reflects one minor, non-load-bearing self-citation.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The paper is a survey; its central thesis depends on unverified domain assumptions about aviation infrastructure, data-sensitivity constraints, and the accuracy of its cited sources, rather than on fitted parameters or new entities.

axioms (3)
  • domain assumption Edge AI models can run on certified avionics-grade hardware and airport edge servers with sufficient performance.
    Stated in Section V-A as a vision ('models can execute on avionics-grade hardware or airport edge servers') and implicitly assumed by the deployment tables in Section IV-A; no certification evidence is supplied.
  • domain assumption Raw aviation data centralization is a binding privacy and sovereignty risk for cloud AI.
    Introduction and Section II-C assert privacy and security as primary motivations without quantifying the risk or comparing with existing regulatory frameworks.
  • domain assumption The cited sources accurately represent the capabilities of edge-AI techniques.
    The review's technical claims about quantization, pruning, distillation, federated learning, and split learning are inherited from cited papers; the paper does not independently verify them.

pith-pipeline@v1.3.0-alltime-deepseek · 12562 in / 9851 out tokens · 101364 ms · 2026-08-01T12:00:13.477465+00:00 · methodology

0 comments
read the original abstract

Civil aviation is safety critical and its operations, from flight decks and towers to ramps and maintenance, generate massive, heterogeneous data at the network edge. Yet cloud centric deployment of large Artificial Intelligence (AI) models often produces high task latency, lacks offline capability in communication denied environments, and requires centralizing sensitive data, raising privacy and sovereignty risks. Edge AI moves perception, prediction, and decision logic closer to the data producers via compression, collaborative inference, and split learning, thereby reducing latency, bandwidth, and exposure while enabling graceful operation during disconnections. This paper provides a panoramic view and a common understanding of edge intelligence tailored to civil aviation. We firstly articulate the operational motivations for edge AI, and then review recent techniques for edge inference and edge learning. We then introduce the organizational computing paradigms and the respective configurations in civil aviation environments; finally, we describe the emerging applications and the future research trends of edge intelligence in civil aviation. We argue that a refined edge solution can complement cloud foundations to deliver low latency, privacy preserving, and resilient AI services across the civil aviation lifecycle.

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

53 extracted references · 15 canonical work pages · 4 internal anchors

  1. [1]

    Artificial intelligence in aviation safety: Systematic review and biometric analysis,

    G. Demir, S. Moslem, and S. Duleba, “ Artificial intelligence in aviation safety: Systematic review and biometric analysis, ” Int. J. Comput. Intell. Syst., vol. 17, no. 1, p. 279, Nov. 2024, doi:10.1007/s44196-024-00671-w

    work page doi:10.1007/s44196-024-00671-w 2024

  2. [2]

    A comprehensive review and a taxonomy of edge machine learning: Requirements, paradigms, and techniques,

    W. Li, H. Hacid, E. Almazrouei, and M. Debbah, “A comprehensive review and a taxonomy of edge machine learning: Requirements, paradigms, and techniques,” AI, vol. 4, no. 3, pp. 729 –786, Sep. 2023, doi:10.3390/ai4030039

    work page doi:10.3390/ai4030039 2023

  3. [3]

    Edge computing for aviation autonomy,

    M. Abdelbaky, J. Chen, A. K. Ishihara, C. Joe-Wong, and S. D. Shetye, “Edge computing for aviation autonomy, ” KBR Tech. Rep., Sep

  4. [4]

    A survey on collaborative DNN inference for edge intelligence,

    W.-Q. Ren, Y. -B. Qu, C. Dong, Y. -Q. Jing, H. Sun, Q. -H. Wu, and S. Guo, “A survey on collaborative DNN inference for edge intelligence, ” Int. J. Autom. Comput., vol. 20, no. 3, pp. 370 –395, May 2023, doi:10.1007/s11633-022-1391-7

    work page doi:10.1007/s11633-022-1391-7 2023

  5. [5]

    Edge intelligence: The confluence of edge computing and artificial intelligence,

    S. Deng, H. Zhao, W. Fang, J. Yin, S. Dustdar, and A. Y. Zomaya, “ Edge intelligence: The confluence of edge computing and artificial intelligence,” IEEE Internet Things J., vol. 7, no. 8, pp. 7457 –7469, Aug. 2020, doi:10.1109/JIOT.2020.2984887

    arXiv 2020

  6. [6]

    Mobile edge intelligence for large language models: A contemporary survey,

    “Mobile edge intelligence for large language models: A contemporary survey, ” IEEE Xplore, [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10835069/citations#citatio ns

    arXiv

  7. [7]

    Edge intelligence for smart airport runway: Architectures and enabling technologies,

    X. Liu, Q. Wang, C. Zou, M. Yu, and D. Liao, “Edge intelligence for smart airport runway: Architectures and enabling technologies, ” Comput. Commun., vol. 195, pp. 323 – 333, Nov. 2022, doi:10.1016/j.comcom.2022.09.003

    work page doi:10.1016/j.comcom.2022.09.003 2022

  8. [8]

    How edge computing can bolster aviation sector innovation,

    “ How edge computing can bolster aviation sector innovation, ” Information Age, [Online]. Available: https://www.information - age.com/how-edge-computing-can-bolster-aviation-sector-innovation- 19833/

  9. [9]

    Phi-3 technical report: A highly capable language model locally on your phone,

    M. Abdin, J. Aneja, H. Awadalla, et al., “Phi-3 technical report: A highly capable language model locally on your phone, ” arXiv:2404.14219, 2024, doi:10.48550/arXiv.2404.14219

    Pith/arXiv arXiv doi:10.48550/arxiv.2404.14219 2024

  10. [10]

    SmoothQuant: Accurate and efficient post-training quantization for large language models,

    G. Xiao, J. Lin, M. Seznec, H. Wu, J. Demouth, and S. Han, “ SmoothQuant: Accurate and efficient post-training quantization for large language models, ” arXiv:2211.10438, 2024, doi:10.48550/arXiv.2211.10438

    Pith/arXiv arXiv doi:10.48550/arxiv.2211.10438 2024

  11. [11]

    QLoRA: Efficient finetuning of quantized LLMs,

    T. Dettmers, A. Pagnoni, A. Holtzman, and L. Zettlemoyer, “QLoRA: Efficient finetuning of quantized LLMs, ” arXiv:2305.14314, 2023, doi:10.48550/arXiv.2305.14314

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.14314 2023

  12. [12]

    AWQ: Activation -aware weight quantization for on -device LLM compression and acceleration,

    J. Lin, J. Tang, H. Tang, S. Yang, W.-M. Chen, W.-C. Wang, G. Xiao, X. Dang, C. Gan, and S. Han, “ AWQ: Activation -aware weight quantization for on -device LLM compression and acceleration, ” Proc. Mach. Learn. Syst., vol. 6, pp. 87–100, May 2024

    2024

  13. [13]

    AWQ: Activation -aware weight quantization for LLM compression and acceleration,

    J. Lin, J. Tang, H. Tang, S. Yang, W.-M. Chen, W.-C. Wang, G. Xiao, X. Dang, C. Gan, and S. Han, “ AWQ: Activation -aware weight quantization for LLM compression and acceleration, ” arXiv:2306.00978, 2024, doi:10.48550/arXiv.2306.00978

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.00978 2024

  14. [14]

    SqueezeLLM: Dense -and-sparse quantization,

    S. Kim, C. Hooper, A. Gholami, Z. Dong, X. Li, S. Shen, M. W. Mahoney, and K. Keutzer, “ SqueezeLLM: Dense -and-sparse quantization, ” arXiv:2306.07629, 2024, doi:10.48550/arXiv.2306.07629

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.07629 2024

  15. [15]

    SparseGPT: Massive language models can be accurately pruned in one -shot,

    E. Frantar and D. Alistarh, “SparseGPT: Massive language models can be accurately pruned in one -shot, ” arXiv:2301.00774, 2023, doi:10.48550/arXiv.2301.00774

    Pith/arXiv arXiv doi:10.48550/arxiv.2301.00774 2023

  16. [16]

    A simple and effective pruning approach for large language models,

    M. Sun, Z. Liu, A. Bair, and J. Z. Kolter, “A simple and effective pruning approach for large language models,” arXiv:2306.11695, 2024, doi:10.48550/arXiv.2306.11695

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.11695 2024

  17. [17]

    Fluctuation-based adaptive structured pruning for large language models,

    Y. An, X. Zhao, T. Yu, M. Tang, and J. Wang, “Fluctuation-based adaptive structured pruning for large language models, ” arXiv:2312.11983, 2023, doi:10.48550/arXiv.2312.11983

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.11983 2023

  18. [18]

    LLM pruning and distillation in practice: The Minitron approach,

    S. T. Sreenivas, S. Muralidharan, R. Joshi, et al., “LLM pruning and distillation in practice: The Minitron approach, ” arXiv:2408.11796, 2024, doi:10.48550/arXiv.2408.11796

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.11796 2024

  19. [19]

    Distil-Whisper: Robust knowledge distillation via large -scale pseudo labelling,

    S. Gandhi, P. von Platen, and A. M. Rush, “Distil-Whisper: Robust knowledge distillation via large -scale pseudo labelling, ” arXiv:2311.00430, 2023, doi:10.48550/arXiv.2311.00430

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.00430 2023

  20. [20]

    A survey on knowledge distillation of large language models,

    X. Xu, M. Li, C. Tao, T. Shen, R. Cheng, J. Li, C. Xu, D. Tao, and T. Zhou, “A survey on knowledge distillation of large language models,” arXiv:2402.13116, 2024, doi:10.48550/arXiv.2402.13116

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.13116 2024

  21. [21]

    Survey on knowledge distillation for large language models: Methods, evaluation, and application,

    C. Yang, Y. Zhu, W. Lu, Y. Wang, Q. Chen, C. Gao, B. Yan, and Y. Chen, “Survey on knowledge distillation for large language models: Methods, evaluation, and application, ” ACM Trans. Intell. Syst. Technol., Oct. 2024, doi:10.1145/3699518

    doi:10.1145/3699518 2024

  22. [22]

    DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning,

    DeepSeek-AI, D. Guo, D. Yang, et al., “DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning, ” arXiv:2501.12948, 2025, doi:10.48550/arXiv.2501.12948

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.12948 2025

  23. [23]

    LLaMA-Adapter: Efficient fine -tuning of language models with zero -init attention,

    R. Zhang, J. Han, C. Liu, P. Gao, A. Zhou, X. Hu, S. Yan, P. Lu, H. Li, and Y. Qiao, “ LLaMA-Adapter: Efficient fine -tuning of language models with zero -init attention, ” arXiv:2303.16199, 2024, doi:10.48550/arXiv.2303.16199

    Pith/arXiv arXiv doi:10.48550/arxiv.2303.16199 2024

  24. [24]

    Split computing and early exiting for deep learning applications: Survey and research challenges,

    Y. Matsubara, M. Levorato, and F. Restuccia, “Split computing and early exiting for deep learning applications: Survey and research challenges,” ACM Comput. Surv., vol. 55, no. 5, pp. 90:1 –90:30, Dec. 2022, doi:10.1145/3527155

    doi:10.1145/3527155 2022

  25. [25]

    Communication-efficient learning of deep networks from decentralized data,

    “Communication-efficient learning of deep networks from decentralized data, ” Proc. Mach. Learn. Res., [Online]. Available: https://proceedings.mlr.press/v54/mcmahan17a.html

  26. [26]

    A survey on efficient federated learning methods for foundation model training,

    “A survey on efficient federated learning methods for foundation model training, ” Proc. IJCAI, [Online]. Available: https://www.ijcai.org/proceedings/2024/919

    2024

  27. [27]

    Federated learning survey: A multi -level taxonomy of aggregation techniques, experimental insights, and future frontiers,

    M. Arbaoui, M.-e.-A. Brahmia, A. Rahmoun, and M. Zghal, “Federated learning survey: A multi -level taxonomy of aggregation techniques, experimental insights, and future frontiers, ” ACM Trans. Intell. Syst. Technol., vol. 15, no. 6, pp. 113:1 – 113:69, Nov. 2024, doi:10.1145/3678182

    work page doi:10.1145/3678182 2024

  28. [28]

    Recent Advances on Federated Learning: A Systematic Survey

    B. Liu, N. Lv, Y. Guo, and Y. Li, “Recent advances on federated learning: A systematic survey, ” arXiv:2301.01299, 2023, doi:10.48550/arXiv.2301.01299

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2301.01299 2023

  29. [29]

    Federated learning based on Stackelberg game in unmanned -aerial-vehicle-enabled mobile edge computing,

    C. Li, M. Song, and Y. Luo, “Federated learning based on Stackelberg game in unmanned -aerial-vehicle-enabled mobile edge computing, ” Expert Syst. Appl., vol. 235, p. 121023, Jan. 2024, doi:10.1016/j.eswa.2023.121023

    arXiv 2024

  30. [30]

    Split learning for health: Distributed deep learning without sharing raw patient data,

    P. Vepakomma, O. Gupta, T. Swedish, and R. Raskar, “Split learning for health: Distributed deep learning without sharing raw patient data, ” arXiv:1812.00564, 2018, doi:10.48550/arXiv.1812.00564

    Pith/arXiv arXiv doi:10.48550/arxiv.1812.00564 2018

  31. [31]

    Edge-MSL: Split learning on the mobile edge via multi -armed bandits,

    T. Kim, J. Zuo, X. Zhang, and C. Joe-Wong, “Edge-MSL: Split learning on the mobile edge via multi -armed bandits, ” in Proc. IEEE INFOCOM, Vancouver, BC, Canada, May 2024, pp. 391 – 400, doi:10.1109/INFOCOM52122.2024.10621231

    arXiv 2024

  32. [32]

    Security Analysis of SplitFed Learning

    M. A. Khan, V. Shejwalkar, A. Houmansadr, and F. M. Anwar, “ Security analysis of SplitFed learning, ” arXiv:2212.01716, 2022, doi:10.48550/arXiv.2212.01716

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2212.01716 2022

  33. [33]

    ATCO2 corpus: A large-scale dataset for research on automatic speech recognition and natural language understanding of air traffic control communications,

    J. Zuluaga-Gomez, K. Veselý, I. Szöke, A. Blatt, P. Motlicek, M. Kocour, M. Rigault, K. Choukri, A. Prasad, S. S. Sarfjoo, I. Nigmatulina, C. Cevenini, P. Kolčárek, A. Tart, J. Černocký, and D. Klakow, “ATCO2 corpus: A large-scale dataset for research on automatic speech recognition and natural language understanding of air traffic control communications,...

    Pith/arXiv arXiv doi:10.48550/arxiv.2211.04054 2023

  34. [34]

    Lessons learned in transcribing 5000 h of air traffic control communications for robust automatic speech understanding,

    J. Zuluaga-Gomez, I. Nigmatulina, A. Prasad, P. Motlicek, D. Khalil, S. Madikeri, A. Tart, I. Szoke, V. Lenders, M. Rigault, and K. Choukri, “ Lessons learned in transcribing 5000 h of air traffic control communications for robust automatic speech understanding, ” Aerospace, vol. 10, no. 10, p. 898, Oct. 2023, doi:10.3390/aerospace10100898

    work page doi:10.3390/aerospace10100898 2023

  35. [35]

    A surveillance video real-time object detection system based on edge- cloud cooperation in airport apron,

    “A surveillance video real-time object detection system based on edge- cloud cooperation in airport apron, ” MDPI Sensors, [Online]. Available: https://www.mdpi.com/2076 -3417/12/19/10128. [Accessed: 4-Sep-2025]

    2076

  36. [36]

    Intelligent surveillance of airport apron: Detection and location of abnormal behavior in typical non -cooperative human objects,

    “Intelligent surveillance of airport apron: Detection and location of abnormal behavior in typical non -cooperative human objects, ” MDPI Sensors, [Online]. Available: https://www.mdpi.com/2076 - 3417/14/14/6182

    2076

  37. [37]

    Edge-cloud collaborative streaming video analytics with multi-agent deep reinforcement learning,

    B. Qian, Y. Xuan, D. Wu, Z. Wen, R. Yang, S. He, J. Chen, and R. Ranjan, “ Edge-cloud collaborative streaming video analytics with multi-agent deep reinforcement learning, ” IEEE Netw., vol. 39, no. 1, pp. 165–173, Jan. 2025, doi:10.1109/MNET.2024.3398724

    arXiv 2025

  38. [38]

    Large-scale video analytics with cloud – edge collaborative continuous learning,

    Y. Nan, S. Jiang, and M. Li, “Large-scale video analytics with cloud – edge collaborative continuous learning,” ACM Trans. Sens. Netw., vol. 20, no. 1, pp. 14:1–14:23, Oct. 2023, doi:10.1145/3624478

    work page doi:10.1145/3624478 2023

  39. [39]

    Application of physical - structure-driven deep learning and compensation methods in aircraft engine health management,

    D. Xiao, H. Xiao, R. Li, and Z. Wang, “Application of physical - structure-driven deep learning and compensation methods in aircraft engine health management, ” Eng. Appl. Artif. Intell., vol. 136, p. 109024, Oct. 2024, doi:10.1016/j.engappai.2024.109024

    arXiv 2024

  40. [40]

    A health state -related ensemble deep learning method for aircraft engine remaining useful life prediction,

    “ A health state -related ensemble deep learning method for aircraft engine remaining useful life prediction, ” Appl. Soft Comput., Aug. 2025, doi:10.1016/j.asoc.2023.110041

    arXiv 2025

  41. [41]

    Predictive maintenance analytics and implementation for aircraft: Challenges and opportunities,

    I. Stanton, K. Munir, A. Ikram, and M. El -Bakry, “ Predictive maintenance analytics and implementation for aircraft: Challenges and opportunities, ” Syst. Eng., vol. 26, no. 2, pp. 216 – 237, 2023, doi:10.1002/sys.21651

    work page doi:10.1002/sys.21651 2023

  42. [42]

    A computer vision-based standalone system for automated operational data collection at non-towered airports,

    M. Farhadmanesh, A. Rashidi, A. K. Subedi, and N. Marković, “A computer vision-based standalone system for automated operational data collection at non-towered airports,” IEEE Access, vol. 12, pp. 95245 – 95264, Jan. 2024, doi:10.1109/ACCESS.2024.3425572

    arXiv 2024

  43. [43]

    A survey on UAV-enabled edge computing: Resource management perspective,

    X. Xia, S. M. M. Fattah, and M. A. Babar, “A survey on UAV-enabled edge computing: Resource management perspective, ” ACM Comput. Surv., vol. 56, no. 3, pp. 78:1–78:36, Oct. 2023, doi:10.1145/3626566

    work page doi:10.1145/3626566 2023

  44. [44]

    A Comprehensive Survey on Aerial Mobile Edge Computing: Challenges, State-of-the-Art, and Future Directions

    Z. Song, X. Qin, Y. Hao, T. Hou, J. Wang, and X. Sun, “ A comprehensive survey on aerial mobile edge computing: Challenges, state-of-the-art, and future directions, ” arXiv:2208.13965, 2022, doi:10.48550/arXiv.2208.13965

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2208.13965 2022

  45. [45]

    Conformal alignment: Knowing when to trust foundation models with guarantees,

    Y. Gui, Y. Jin, and Z. Ren, “Conformal alignment: Knowing when to trust foundation models with guarantees, ” Adv. Neural Inf. Process. Syst., vol. 37, pp. 73884–73919, Dec. 2024

    2024

  46. [46]

    Deep correlated prompting for visual recognition with missing modalities,

    L. Hu, T. Shi, W. Feng, F. Shang, and L. Wan, “ Deep correlated prompting for visual recognition with missing modalities, ” arXiv:2410.06558, 2024, doi:10.48550/arXiv.2410.06558

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.06558 2024

  47. [47]

    A survey on efficient federated learning methods for foundation model training,

    H. Woisetschläger, A. Erben, S. Wang, R. Mayer, and H. -A. Jacobsen, “A survey on efficient federated learning methods for foundation model training, ” in Proc. IJCAI, 2024, pp. 8317 – 8325, doi:10.24963/ijcai.2024/919

    work page doi:10.24963/ijcai.2024/919 2024

  48. [48]

    Task-oriented energy scheduling in wireless rechargeable sensor networks,

    J. Zhang, H. Gao, Q. Chen, and J. Li, “Task-oriented energy scheduling in wireless rechargeable sensor networks, ” ACM Trans. Sens. Netw., vol. 19, no. 4, pp. 88:1–88:32, Jun. 2023, doi:10.1145/3594874

    work page doi:10.1145/3594874 2023

  49. [49]

    HEMM: Holistic Evaluation of Multimodal Foundation Models

    P. P. Liang, A. Goindani, T. Chafekar, L. Mathur, H. Yu, R. Salakhutdinov, and L. -P. Morency, “HEMM: Holistic evaluation of multimodal foundation models, ” arXiv:2407.03418, 2024, doi:10.48550/arXiv.2407.03418

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2407.03418 2024

  50. [50]

    Deep multimodal learning with missing modality: A survey,

    R. Wu, H. Wang, H. -T. Chen, and G. Carneiro, “Deep multimodal learning with missing modality: A survey, ” arXiv:2409.07825, 2024, doi:10.48550/arXiv.2409.07825

    Pith/arXiv arXiv doi:10.48550/arxiv.2409.07825 2024

  51. [51]

    TerraMind: Large -scale generative multimodality for earth observation,

    “ TerraMind: Large -scale generative multimodality for earth observation, ” arXiv:2504.11171v3, [Online]. Available: https://arxiv.org/html/2504.11171v3

    Pith/arXiv arXiv

  52. [52]

    AirTrafficGen: Configurable air traffic scenario generation with large language models,

    D. S. W. Gould, G. De Ath, B. Carvell, and N. Pepper, “AirTrafficGen: Configurable air traffic scenario generation with large language models, ” arXiv:2508.02269, 2025, doi:10.48550/arXiv.2508.02269

    Pith/arXiv arXiv doi:10.48550/arxiv.2508.02269 2025

  53. [2023]

    Available: https://www.kbr.com/sites/default/files/documents/2023-09/aaiml- 7507.pdf

    [Online]. Available: https://www.kbr.com/sites/default/files/documents/2023-09/aaiml- 7507.pdf

    2023