The DOI matches its Crossref record; 280 of the 364 references it cites were found in a registry or an index, and 1 are not checked yet; 35 of the 269 sources its claims cite were checked for support.
Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications
Open accessAbstract
The thriving of artificial intelligence (AI) applications is driving the further evolution of wireless networks. It has been envisioned that 6G will be transformative and will revolutionize the evolution of wireless from “connected things” to “connected intelligence”. However, state-of-the-art deep learning and big data analytics based AI systems require tremendous computation and communication resources, causing significant latency, energy consumption, network congestion, and privacy leakage in both of the training and inference processes. By embedding model training and inference capabilities into the network edge, edge AI stands out as a disruptive technology for 6G to seamlessly integrate sensing, communication, computation, and intelligence, thereby improving the efficiency, effectiveness, privacy, and security of 6G networks. In this paper, we shall provide our vision for scalable and trustworthy edge AI systems with integrated design of wireless communication strategies and decentralized machine learning models. New design principles of wireless networks, service-driven resource allocation optimization methods, as well as a holistic end-to-end system architecture to support edge AI will be described. Standardization, software and hardware platforms, and application scenarios are also discussed to facilitate the industrialization and commercialization of edge AI systems.
Verified 2026-10-05 · 1 reference not checked yet
Overview
CoChat Summary
TL;DR
This paper presents a vision for edge AI in 6G, arguing that connected intelligence requires co-design of wireless communication, decentralized machine learning, and resource allocation. It surveys enabling technologies, an end-to-end architecture, standardization/platforms, and applications for scalable and trustworthy edge AI.
Key findings
- Edge AI is positioned as a disruptive 6G technology that embeds model training and inference at the network edge to improve efficiency, effectiveness, privacy, and security; the main challenges identified are trustworthiness and scalability.
- Communication-efficient edge training is supported by over-the-air computation (AirComp), grant-free random access, NOMA, massive MIMO/cell-free massive MIMO, RIS, and SAGIN/UAV networks; edge inference is categorized as horizontal (edge device distributed inference, edge server cooperative inference) and vertical (edge device-server co-inference, URLLC, task-oriented communication).
- Resource allocation for edge AI is framed as service-driven optimization, categorized into mixed-combinatorial, nonconvex, stochastic, and end-to-end optimization, with a “learning to optimize” framework using imitation learning, algorithm unrolling, graph neural networks, and reinforcement learning.
- Applications and standardization discussed include autonomous driving, IoT/AIoT/IIoT, and smart healthcare; the paper notes IEEE 3652.1-2020 for federated machine learning was approved in March 2021, 3GPP may start 6G study around end of 2025 (Release 20) and technical specifications around end of 2027, THz band 0.1–10 THz, AR/VR latency requirements around 100 ms, resource allocation inference within channel coherence time around 10 ms, and AlphaGo may cost 280 GPUs and a $3000 electric bill per game.
Methods
- This is a vision/survey/tutorial paper, not an empirical study; it has no experimental design, dataset, sample, or quantitative evaluation.
- It synthesizes literature and proposes taxonomies, design principles, and an end-to-end architecture across edge training, edge inference, resource allocation, standardization, platforms, and applications.
- The main analysis is conceptual and technical synthesis, supported by figures, a summary table, and references to prior work rather than new experiments or meta-analysis.
Limitations
- The paper states some technology limitations: building an edge AI management/orchestration framework that fully relies on standardization may not be feasible; more efforts are needed to incorporate AirComp functionalities into future 6G standards; and finding the globally optimal solution for distributionally robust optimization is often computationally intractable.
- Evident from the text: the paper reports no experiments, simulations, measured performance, datasets, or sample sizes, so its claims are visionary and broad rather than empirically validated.
- Evident from the text: there is no explicit limitations section for the paper itself, and it does not discuss validation or generalizability of the proposed architecture.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
28 Valid · 1 Not valid · 155 Not checked · 36 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Inspired by the dynamic spiking neurons in the human brain, the energy consumption and latency of edge AI can be significantly reduced by processing the learning tasks in an event-driven manner [62], [63].”
§ Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications > C. Edge AI Empowered 6G Networks
What the sources say
- Partly supportedcites [62] sourceExperimental — accuracy not yet measured
The Introduction (pp. 1–2) links brain-inspired SNNs to efficient online learning on low-power devices and reports substantial energy-efficiency gains. The Conclusions (p. 21) describe event-driven learning and inference, but characterize delay as part of an accuracy–energy–delay tradeoff; they do not establish a significant reduction in latency for edge AI. (LLM, conf=0.86)
- Partly supportedcites [63] sourceExperimental — accuracy not yet measured
The Introduction (page not available in our copy) describes spiking neural networks as a low-power alternative for battery-powered devices because they process information in an event-driven, sparse, online fashion. It does not establish that this processing significantly reduces latency, so the sentence's latency claim is unsupported by this source. (LLM, conf=0.87)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“For instance, edge DL approach has been developed in [58] to deliver low-latency semantic tasks (e.g., text messages) by learning the communication strategies in an end-to-end fashion based on JSCC.”
§ Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications > C. Edge AI Empowered 6G Networks
What the sources say
- Partly supportedcites [58] sourceExperimental — accuracy not yet measured
The Introduction (p. 3) describes a distributed semantic communication network designed under latency constraints, with jointly optimized receiver and feature-extractor networks. The Semantic Communication System section (p. 3) describes a text transmitter and receiver with semantic and channel encoders and decoders, supporting the end-to-end coding characterization. The paper does not demonstrate low-latency delivery of semantic tasks; its reported latency concern is model broadcasting, and its runtime comparison is not an end-to-end task-latency measurement. (LLM, conf=0.82)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“For instances, distributed RL [55], decentralized graph neural networks [23], and distributed DNN [61], are able to automatically learn the distributed resource allocation optimization algorithms.”
§ Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications > C. Edge AI Empowered 6G Networks
What the sources say
- Supportedcites [55] source
The citation to this paper, [55], supports the distributed RL example, not the separately cited graph-neural-network or DNN examples. The Introduction (p. 1) says its distributively executed deep-Q-learning algorithm learns a policy for transmitters to adjust their power levels; Section B (p. 5) clarifies that execution is distributed while training is centralized. (LLM, conf=0.91)
- Not checked yetcites [23] source
- Supportedcites [61] source
The citation to this paper is attached to “distributed DNN,” not to distributed RL or graph neural networks. Section IV describes jointly trained DNNs that approximate a distributed optimizer, and the Conclusions state that the framework learns unknown solution rules and was tested on wireless resource-allocation problems. (LLM, conf=0.91)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“In particular, swarm learning [35] provides a completely decentralized AI solution based on decentralized ML by keeping local datasets at each edge device.”
§ Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications > A. Edge Learning Models and Algorithms
What the sources say
- Supportedcites [35] source
The Main section states the hypothesis of 'completely decentralized AI solutions' whose first property is that it 'keeps large medical data locally with the data owner' with no exchange of raw data (Main, p. 1). The Concept of Swarm Learning section adds that SL 'dispenses with a dedicated server' and 'builds the models independently on private data at the individual sites (short ‘nodes’ called Swarm edge nodes)' (Concept of Swarm Learning, p. 2), and the Methods describe the SLL as enabling 'decentralized training of machine learning models without sharing the data' (Methods, The Swarm Learning framework). Together these support both parts of the claim: decentralized ML and retention of local data at each edge node. (LLM, conf=0.92)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“It can be typically applied to the medical diagnosis and millimeter wave channel prediction [37].”
§ Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications > A. Edge Learning Models and Algorithms
What the sources say
- Supportedcites [37] source
Section 3.4 Split Learning (p. 4) states that SL 'has recently been adopted in medical applications wherein dispersed private health records should be exploited without sharing raw data' and 'has also been known for its robustness against non-IID data distributions, and applied for fusing heterogeneous vision and radio-frequency (RF) modalities to predict millimeter-wave channels.' The paper further develops exactly these two application areas as use cases in Sec. 6.10 (Tripartite SL for Medical Diagnosis, pp. 13-14) and Sec. 6.12 (Heteromodal SL for mmWave Channel Prediction, pp. 15-16), so the two exemplars attributed to split learning are the source's own characterizations. (LLM, conf=0.90)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
136 more claims · 16 with their cited sources checked
Create a free account to see all 220 claims and to verify the 234 references still open.
- Probably not a substantial claim. 48
- It may state two separate claims as one. 36
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“With the right combination of hardware and software, Edge AI technology can enable AI algorithms to run on edge devices, bringing intelligence to the network's edge and enabling new applications and use cases [32].”
From Edge AI: A survey (2023) · cites this paper as [32]
Citations
Cites
A Mathematical Theory of Communication
Shannon, C. E. (1948) · Bell System Technical Journal · doi:10.1002/j.1538-7305.1948.tb00917.x · Cited by 48,949
MapReduce: simplified data processing on large clusters
Dean, J., & Ghemawat, S. (2008) · Communications of the ACM · doi:10.1145/1327452.1327492 · Cited by 18,476 · cited 1×
“communication among edge devices, thereby constructing the desired computation or inference results using the reduce function [222], [52].”
MethodAndrews, J. G., Buzzi, S., Choi, W., Hanly, S. V., Lozano, A., Soong, A. C. K., & Zhang, J. C. (2014) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2014.2328098 · Cited by 8,106 · cited 1×
“For example, based on typical use cases in 5G [5], [6] (i.e., enhanced mobile broadband (eMBB), ultra-reliable and low-latency communications (URLLC), and massive machine type communications (mMTC)), Huawei has recently proposed three additional application scenarios in the vision of 5.5G.”
BackgroundAdvances and Open Problems in Federated Learning
Kairouz, P., & McMahan, H. B. (2021) · Foundations and Trends in Machine Learning · doi:10.1561/2200000083 · Cited by 7,772 · cited 3×
“Section II: CommunicationEfficient Edge Training Section II-A: Edge Learning Models and Algorithms Federated Learning Horizontal and vertical federated learning [67], [34], [33]; federated optimization [68]”
BackgroundMethodAsynchronous Methods for Deep Reinforcement Learning
Mnih, V., Badia, A. P., Mirza, M., Graves, A., Lillicrap, T. P., Harley, T., Silver, D., & Kavukcuoglu, K. (2016) · arXiv · doi:10.48550/arxiv.1602.01783 · Cited by 7,132
Federated Machine Learning: Concept and Applications
Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2019) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3298981 · Cited by 6,322
Federated learning: Challenges, methods, and future directions
Li, T., Sahu, A. K., Talwalkar, A., & Smith, V. (2020) · IEEE Signal Processing Magazine · doi:10.1109/msp.2020.2975749 · Cited by 6,210 · cited 1×
“…communication costs with a large model frequently exchanged over wireless networks, statistical heterogeneity with non-identical local data distributions and sizes, system heterogeneity with varied storage, computation and communication capabilities, as well as dynamic devices participation [122].”
BackgroundDigital Twin in Industry: State-of-the-Art
Tao, F., Zhang, H., Liu, A., & Nee, A. Y. C. (2019) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2018.2873186 · Cited by 5,871 · cited 1×
“Besides, digital twin is becoming a key technology for smart manufacturing in industrial 4.0 by connecting physical machines and digital representations in a cyber-physical system [308], [309].”
MethodInfluentialA Vision of 6G Wireless Systems: Applications, Trends, Technologies, and Open Research Problems
Saad, W., Bennis, M., & Chen, M. (2020) · IEEE Network · doi:10.1109/mnet.001.1900287 · Cited by 4,985 · cited 1×
“…haptic internet, high-fidelity holographic society, extended reality (XR) and metaverse [9], e-health, etc. Researchers in industry and academia have published many visionary 6G proposals [10], [11], [12] to provide a better understanding, sensing, controlling, and interacting for a physical world.”
BackgroundDeep Reinforcement Learning: A Brief Survey
Arulkumaran, K., Deisenroth, M. P., Brundage, M., & Bharath, A. A. (2017) · IEEE Signal Processing Magazine · doi:10.1109/msp.2017.2743240 · Cited by 4,733 · cited 1×
“This can be frequently modeled as decision making and learning in a Markov decision process (MDP) [160].”
Background
Research works printed but not linked: 49
- Expanded 6G Vision, Use Cases and Societal Values
- IMT-2030 (6G) Promotion Group, 6G Vision and Candidate Technologies
- 2019, Network 2030: A Blueprint of Technology, Applications and Market Drivers Towards the Year 2030 and Beyond
- Huawei 5.5G
- Ali, 2020, White paper on machine learning in wireless communication networks, 6G Res. Vis., 1
- Kim, Communication algorithms via deep learning, 1
- 10.1109/JPROC.2021.3055679 · DOI · Paper page
- Zhang, Fully decentralized multi-agent reinforcement learning with networked agents, 5872
- McMahan, Communication-efficient learning of deep networks from decentralized data, 54, 1273
- Lu, Optimal complexity in decentralized training, 7111
- Li, Communication efficient distributed machine learning with the parameter server, 27, 19
- 2020, Harmonizing Standards for Edge Computing
- Fate
- HarmonyOS
- Alistarh, QSGD: Communication-efficient SGD via gradient quantization and encoding, 1707
- 10.5555/3294771.3294915 · doi:10.5555/3294771.3294915
- Bernstein, signSGD: Compressed optimisation for non-convex problems, 80, 560
- Wang, ATOMO: Communication-efficient learning via atomic sparsification, 9872
- Liu, 2021, Hierarchical quantized federated learning: Convergence analysis and system design, arXiv:2103.14272 · arXiv
- Haddadpour, Federated learning with compression: Unified analysis and sharp guarantees, 2350
- Li, On the convergence of FedAvg on non-IID data, 1
- Acar, Federated learning based on dynamic regularization, 1
- Dinh, Personalized federated learning with Moreau envelopes, 33, 21394
- Deng, Distributionally robust federated averaging, 33, 15111
- 10.1561/9781680837735 · DOI
- Smith, Federated multi-task learning, 30, 4427
- Fallah, Personalized federated learning with theoretical guarantees: A model-agnostic meta-learning approach, 33, 3557
- Ruan, Towards flexible device participation in federated learning, 130, 3403
- 10.5555/3295222.3295285 · doi:10.5555/3295222.3295285
- Wang, 2018, Cooperative SGD: A unified framework for the design and analysis of communication-efficient SGD algorithms, arXiv:1808.07576 · arXiv
- Koloskova, A unified theory of decentralized SGD with changing topology and local updates, 5381
- Koloskova, Decentralized stochastic optimization and gossip algorithms with compressed communication, 3478
- Neglia, Decentralized gradient methods: Does topology matter?, 108, 2348
- Choromanska, Beyond backprop: Online alternating minimization with auxiliary variables, 97, 1193
- Lowe, Multi-agent actor-critic for mixed cooperative-competitive environments, 6382
- Damaskinos, AGGREGATHOR: Byzantine machine learning via robust gradient aggregation, 1
- Chen, Breaking the communication-privacy-accuracy trilemma, 33, 3312
- Blanchard, Machine learning with adversaries: Byzantine tolerant gradient descent, 118
- Bonawitz, 2019, Towards federated learning at scale: System design, arXiv:1902.01046 · arXiv
- Verma, Manifold mixup: Better representations by interpolating hidden states, 6438
- Tishby, The information bottleneck method, 368
- Jin, How to escape saddle points efficiently, 1724
- 10.1109/TCOMM.2021.3088528 · DOI · Paper page
- 2021, Privacy Protection—Privacy Guidelines for Smart Cities
- 2021, Guidelines for Security and Privacy in Internet of Things
- 2020, Cyber Security for Consumer Internet of Things: Baseline Requirements
- 2018, Technical Specification Group Services and System Aspects; System Architecture for the 5G System; Stage 2 (Release 15)
- Mosek
- Gurobi
Cited by
Semantic Communications for Future Internet: Fundamentals, Applications, and Challenges
Yang, W., Du, H., Liew, Z. Q., Lim, W. Y. B., Xiong, Z., Niyato, D., Chi, X., Shen, X., & Miao, C. (2023) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2022.3223224 · Cited by 670
“Data security and privacy issues are always significant topics in the field of wireless communications [210].”
Xu, W., Yang, Z., Ng, D. W. K., Levorato, M., Eldar, Y. C., & Debbah, M. (2023) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2023.3239189 · Cited by 660
“implement edge artificial intelligence (AI) in B5G networks [8], [9], [10], [11], [12], [13].”
InfluentialSingh, R., & Gill, S. S. (2023) · Internet of Things and Cyber-Physical Systems · doi:10.1016/j.iotcps.2023.02.004 · Cited by 572
Xu, M., Ng, W. C., Lim, W. Y. B., Kang, J., Xiong, Z., Niyato, D., Yang, Q., Shen, X., & Miao, C. (2023) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2022.3221119 · Cited by 564
“towards the co-design of computation and communication systems [24] implies that the design of next-generation mobile edge networks will also attempt to solve the problem of”
BackgroundUnleashing the Power of Edge-Cloud Generative AI in Mobile Networks: A Survey of AIGC Services
Xu, M., Du, H., Niyato, D., Kang, J., Xiong, Z., Mao, S., Han, Z., Jamalipour, A., Kim, D. I., Shen, X., et al. (2024) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2024.3353265 · Cited by 449
“2022 [20] Discusses the visions and potentials of low-power, low-latency, reliable, and trustworthy edge intelligence for 6G wireless networks 7 7 3”
Metadata
- Authors
- Letaief, K. B., Shi, Y., Lu, J., & Lu, J.
- Affiliations
- Hong Kong University of Science and Technology (HK)
- ShanghaiTech University (CN)
- Huawei Technologies (China) (CN)
- Venue
- IEEE Journal on Selected Areas in Communications
- Published
- 2022-01
- Volume
- 40
- Issue
- 1
- Pages
- 5-36
- ISSN
- 0733-8716 (ISSN-L)
- 1558-0008
- OpenAlex
- W3212941463
- Semantic Scholar
- be0bbf06977c4dadbf702287733187884a531b8a
- Cited by
- 829
- References
- 364
- Citations by year
- 2026: 157 · 2025: 213 · 2024: 210 · 2023: 173 · 2022: 75
- FWCI
- 51.30
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- IEEE Journal on Selected Areas in Communications (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- arXiv (Cornell University)
- Funders
- RDC Corporation Ltd. (20210400L016)
- Natural Science Foundation of Shanghai (21ZR1442700)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Wireless Communication Technologies
- Physical Sciences › Computer Science › Computer Networks and Communications › Software-Defined Networks and 5G
- Keywords
- edge AI, 6G, decentralized machine learning, resource allocation optimization, integrated sensing and communication, edge computing, privacy preservation, network security, AI standardization, connected intelligence
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications. https://app.cochat.ai/papers/10.1109/jsac.2021.3126076
- The paper
- Letaief, K. B., Shi, Y., Lu, J., & Lu, J. (2022). Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications. IEEE Journal on Selected Areas in Communications. https://doi.org/10.1109/jsac.2021.3126076
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.