The DOI matches its Crossref record; 128 of the 129 references it cites were found in a registry or an index; 15 of the 54 sources its claims cite were checked for support.
Survey: federated learning data security and privacy-preserving in edge-Internet of Things
Open accessAbstract
The amount of data generated owing to the rapid development of the Smart Internet of Things is increasing exponentially. Traditional machine learning can no longer meet the requirements for training complex models with large amounts of data. Federated learning, as a new paradigm for training statistical models in distributed edge networks, alleviates integration and training problems in the context of massive and heterogeneous data and security protection for private data. Edge computing processes data at the edge layers of data sources to ensure low-data-delay processing; it provides high-bandwidth communication and a stable network environment, and relieves the pressure of processing massive data using a single node in the cloud center. A combination of edge computing and federated learning can further optimize computing, communication, and data security for the edge-Internet of Things. This review investigated the development status of federated learning and expounded on its basic principles. Then, in view of the security attacks and privacy leakage problems of federated learning in the edge Internet of things, relevant work was investigated from cryptographic technologies (such as secure multi-party computation, homomorphic encryption and secret sharing), perturbation schemes (such as differential privacy), adversarial training and other privacy security protection measures. Finally, challenges and future research directions for the integration of edge computing and federated learning are discussed.
Overview
CoChat Summary
## TL;DR
This is a survey of federated learning (FL) data security and privacy-preserving techniques in edge Internet of Things (edge-IoT) environments. It organizes common attacks, reviews defense approaches such as cryptography, differential privacy, adversarial training, blockchain, and knowledge distillation, and outlines open challenges for secure and efficient edge FL.
## Key findings
- The paper identifies major FL data security attacks in edge-IoT as poisoning, sybil, backdoor, Byzantine, free-riding, and adversarial attacks, and privacy leakage attacks as model inversion, refactoring/reconstruction, and model extraction attacks.
- For privacy-preserving FL, it highlights cryptographic methods including secure multi-party computation (SMPC), homomorphic encryption (HE), and secret sharing, noting that SMPC is secure but computationally expensive, while HE generally uses fewer computing resources than SMPC.
- It summarizes differential privacy (DP) as a common protection method and notes that prior work shows privacy can be achieved “with a small additional overhead” when enough devices participate, while maintaining high model accuracy.
- It argues that combining FL with blockchain, asynchronous FL, over-the-air computing, and knowledge distillation can help balance privacy, robustness, communication cost, and latency in edge-IoT settings.
## Methods
- This is a narrative survey, not an original experimental study: it reviews and compares prior FL research for edge-IoT security and privacy.
- The paper organizes the literature into sections on fundamentals, attack types, related works on security and privacy protection, challenges, and future directions.
- Its main analysis is qualitative synthesis of reported methods, advantages, and limitations across defense families such as robust aggregation, client selection, cryptography, DP, adversarial training, blockchain, and KD.
## Limitations
- The paper states that many privacy-preserving methods trade off privacy against training efficiency, and that higher encryption levels increase computational overhead.
- It states that existing FL privacy/security protocols are mainly developed for horizontal FL and may be inefficient or insecure when applied to vertical FL and federated transfer learning.
- Evident from the text: this is a high-level survey and does not report a unified experimental benchmark or a single dataset/sample of its own, so its conclusions depend on the cited literature rather than new empirical evaluation.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures
Verification
Claims the paper makes
11 Valid · 5 Not valid · 22 Not checked · 67 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The bottleneck in the transmission communication retains the training data in edge device nodes to avoid the risk of privacy data leakages to a certain extent (Li et al. 2022a;Ranaweera et al. 2021).”
§ Survey: federated learning data security and privacy-preserving in edge-Internet of Things > 1 Introduction
What the sources say
- Not matchedcites Li et al. 2022a
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Could not verifycites Ranaweera et al. 2021 sourceExperimental — accuracy not yet measured
The Data Privacy and Privacy partitioning sections (pages unavailable) discuss privacy risks in MEC and a trusted local partition for deep-learning tasks. Neither says that training data is retained on edge devices to reduce leakage, or links that practice to a transmission bottleneck. Searches for local training data and the communication bottleneck found no passage establishing the claim. (LLM, conf=0.72)
hedgedstates a possible 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.
- Not matched
Every source the paper cites for this claim was checked and supports it.
In the paper
“(Song et al. 2019) found that AT makes ML models more vulnerable to member inference attacks than those trained using the original training set.”
§ Survey: federated learning data security and privacy-preserving in edge-Internet of Things > 3.2.3 Adversarial training
What the sources say
- Supportedcites Song et al. 2019 source
The Introduction reports that all six adversarial defense methods tested increased membership-inference risk relative to naturally undefended models. This supports the comparison with models trained without adversarial training. (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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“It attempts to arrange all attacks from the training phase of the client, making the FL global model robust to known adversarial attacks (Tramèr et al. 2017).”
§ Survey: federated learning data security and privacy-preserving in edge-Internet of Things > 3.2.3 Adversarial training
What the sources say
- Partly supportedcites Tramèr et al. 2017 sourceExperimental — accuracy not yet measured
Section 3.2 (p. 3) describes adversarial training on adversarial examples, and Appendix C.2 (p. 19) reports improved robustness to the black-box attacks tested for ensemble adversarial training. The paper does not establish robustness for a federated-learning global model or describe arranging all attacks during client training. Section 4.1 (p. 6) also cautions that ordinary adversarial training remains vulnerable to black-box attacks. (LLM, conf=0.88)
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
Every source the paper cites for this claim was checked and supports it.
In the paper
“(Mejia et al. 2019) found that with a model inversion attack, an attacker can use an AT model to generate images that are similar to actual training samples.”
§ Survey: federated learning data security and privacy-preserving in edge-Internet of Things > 3.2.3 Adversarial training
What the sources say
- Supportedcites Mejia et al. 2019 source
The Introduction (p. 2) says adversarially trained models enable better model-inversion reconstructions. Section 4.2 (p. 7) compares those reconstructions with training images and describes matching visual features, supporting the claim that an attacker can generate similar images. (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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This proved that, with sufficient devices participating in federated training, privacy protection could be achieved with a small additional overhead.”
§ Survey: federated learning data security and privacy-preserving in edge-Internet of Things > 3.2.2 Differential privacy technology
What the sources say
- Supportedcites Geyer et al. 2017 source
The Abstract (p. 1) says that with sufficiently many participating clients, the proposed procedure maintains client-level differential privacy at only a minor cost in model performance. The Discussion (p. 6) likewise reports almost no performance cost with 10,000 clients. The supported ‘overhead’ is reduced model performance, not a measured computational or communication overhead. (LLM, conf=0.89)
- Partly supportedcites McMahan et al. 2017 sourceExperimental — accuracy not yet measured
The Abstract (p. 1) says that a sufficiently large number of users enables differential privacy with negligible loss in predictive accuracy. But the Experimental Results (p. 8) say the computation is significantly more expensive; the paper does not support describing the additional computational overhead as small. (LLM, conf=0.89)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
34 more claims · 2 with their cited sources checked
Create a free account to see all 105 claims and to verify the 2 references still open.
- It may state two separate claims as one. 53
- Probably not a substantial claim. 33
- The quote only partly states this claim. 1
Citations
Cites
Communication-Efficient Learning of Deep Networks from Decentralized Data
McMahan, H. B., Moore, E., Ramage, D., Hampson, S., & Arcas, B. A. Y. (2016) · arXiv · doi:10.48550/arxiv.1602.05629 · Cited by 14,215
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204
Explaining and Harnessing Adversarial Examples
Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014) · arXiv · doi:10.48550/arxiv.1412.6572 · Cited by 10,381
Edge computing: Vision and challenges
Shi, W., Cao, J., Zhang, Q., Li, Y., & Xu, L. (2016) · IEEE Internet of Things Journal · doi:10.1109/jiot.2016.2579198 · Cited by 8,733 · cited 1×
“Edge computing (EC) is a new distributed computing paradigm (Shi et al. 2016) for storing, processing, and applying data collected by edge-IoT devices on the edge side close to the data source.”
Douceur, J. R. (2002) · Lecture Notes in Computer Science · doi:10.1007/3-540-45748-8_24 · Cited by 4,419 · cited 1×
“Attackers use these forged identities to control or affect other normal nodes in a distributed network, resulting in network robustness losses (Douceur 2002; Singh and (2006) Eclipse attacks on overlay networks: Threats and defenses.”
Differential Privacy: A Survey of Results
Dwork, C. (2008) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-79228-4_1 · Cited by 3,542 · cited 1×
“(Dwork 2008) proposed the concept of DP in 2006 and provided a rigorous mathematical proof of its security.”
Robust De-anonymization of Large Sparse Datasets
Narayanan, A., & Shmatikov, V. (2008) · 2008 IEEE Symposium on Security and Privacy (sp 2008) · doi:10.1109/sp.2008.33 · Cited by 2,372 · cited 1×
“Examples include malicious speculation by semi-honest adversaries and theft of data by curious attackers (Narayanan and Shmatikov 2008).”
Ensemble Adversarial Training: Attacks and Defenses
Tramèr, F., Kurakin, A., Papernot, N., Goodfellow, I., Boneh, D., & McDaniel, P. (2017) · arXiv · doi:10.48550/arxiv.1705.07204 · Cited by 2,176
Privacy-Preserving Deep Learning via Additively Homomorphic Encryption
Phong, L. T., Aono, Y., Hayashi, T., Wang, L., & Moriai, S. (2018) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2017.2787987 · Cited by 1,662 · cited 1×
“For example, malicious hacker attacks and dishonest participants may use the model parameters to infer the original data (Phong et al. 2017).”
Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning
Hitaj, B., Ateniese, G., & Perez-Cruz, F. (2017) · Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security · doi:10.1145/3133956.3134012 · Cited by 1,377 · cited 1×
“(Hitaj et al. 2017) first proposed a GAN-based data reconstruction attack to steal the private data of model contributors.”
Research works printed but not linked: 8
- Andrew G, Thakkar O, McMahan B, Ramaswamy S (2021) Differentially private learning with adaptive clipping. Adv Neural Inf Process Syst 34:17455–17466
- Arjovsky M, Chintala S, Bottou L (2017) Wasserstein generative adversarial networks. In: Proceedings of the 34th International Conference on Machine Learning, vol 70, pp. 214–223. PMLR, Sydney, NSW, Australia
- Bhagoji AN, Chakraborty S, Mittal P, Calo S (2019) Analyzing federated learning through an adversarial lens. In: Proceedings of the 36th International Conference on Machine Learning, pp. 634–643. PMLR, Long Beach, California, USA
- Bogdanov D, Laur S, Willemson J (2008) Sharemind: A framework for fast privacy-preserving computations. European Symposium on Research in Computer Security. Springer, Berlin, Heidelberg, pp 192–206
- Kilbertus N, Gascón A, Kusner M, Veale M, Gummadi K, Weller A (2018) Blind justice: Fairness with encrypted sensitive attributes. In: International Conference on Machine Learning, pp. 2630–2639. PMLR, Stockholm, Sweden
- Li XW, Chen BH, Yang DQ, Wu GF (2022a) Review of Security Protocols in Edge Computing Environments. J Comp Res Develop 59(4):765–780
- Tramèr F, Zhang F, Juels A, Reiter MK, Ristenpart T (2016) Stealing machine learning models via prediction {APIs}. In: 25th USENIX security symposium (USENIX Security 16), pp. 601–618. USENIX, Austin, TX, USA
- Zhang J, Chen Y, Li H (2022) Privacy Leakage of AT Models in FL Systems. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.108–114. IEEE, New Orleans, LA, USA
Cited by
Navigating the nexus of AI and IoT: A comprehensive review of data analytics and privacy paradigms
Marengo, A. (2024) · Internet of Things · doi:10.1016/j.iot.2024.101318 · Cited by 112
Edge AI for Real-Time Anomaly Detection in Smart Homes
Reis, M. J. C. S., & Serôdio, C. (2025) · Future Internet · doi:10.3390/fi17040179 · Cited by 93
“Federated learning (FL) has emerged as a viable privacy-preserving alternative to cloud-based AI, allowing smart home devices to collaboratively train models without transmitting raw sensor data to a central server [10,11].”
A Survey of Security Strategies in Federated Learning: Defending Models, Data, and Privacy
Manzoor, H. U., Shabbir, A., Chen, A., Flynn, D., & Zoha, A. (2024) · Future Internet · doi:10.3390/fi16100374 · Cited by 58
Federated learning and TinyML on IoT edge devices: Challenges, advances, and future directions
Ramadan, M. N. A., Ali, M. A. H., Khoo, S. Y., & AlKhedher, M. A. (2025) · ICT Express · doi:10.1016/j.icte.2025.06.008 · Cited by 71
A Comprehensive Survey on Generative AI Solutions in IoT Security
López Delgado, J. L., & López Ramos, J. A. (2024) · Electronics · doi:10.3390/electronics13244965 · Cited by 33
Metadata
- Authors
- Li, H., Ge, L., & Tian, L.
- Affiliations
- Guangxi Minzu University (CN)
- Corresponding author
- Ge, Lina
- ORCID
- Ge, Lina
- Venue
- Artificial Intelligence Review
- Published
- 2024-04-29
- Volume
- 57
- Issue
- 5
- ISSN
- 0269-2821 (ISSN-L)
- 1573-7462
- OpenAlex
- W4396238295
- Semantic Scholar
- 71d606deed3187590a9b86060bf793d29d0d060f
- Cited by
- 112
- References
- 107
- Citations by year
- 2026: 62 · 2025: 40 · 2024: 10
- FWCI
- 16.46
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Artificial Intelligence Review (publishedVersion, cc-by)
- Funders
- National Natural Science Foundation of China (61862007)
- Natural Science Foundation of Guangxi Zhuang Autonomous Region (2020GXNSFBA297103)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Computer Networks and Communications › Cryptography and Data Security
- Keywords
- federated learning, edge computing, Internet of Things, data security, privacy preservation, privacy leakage, secure multi-party computation, homomorphic encryption, secret sharing, differential privacy, adversarial training, privacy attacks, smart Internet of Things
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 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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: Survey: federated learning data security and privacy-preserving in edge-Internet of Things. https://app.cochat.ai/papers/10.1007/s10462-024-10774-7
- The paper
- Li, H., Ge, L., & Tian, L. (2024). Survey: federated learning data security and privacy-preserving in edge-Internet of Things. Artificial Intelligence Review. https://doi.org/10.1007/s10462-024-10774-7
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.

