CoChat
Sign Up
Cited by 412Privacy-Preserving Technologies in DataIoT and Edge/Fog ComputingInternet of Things and AI

Federated learning for the Internet of Things: Applications, challenges, and opportunities

Zhang, Tuo, Gao, Lei, He, Chaoyang, Zhang, Mi, Krishnamachari, Bhaskar, Avestimehr, A. Salman·2022·IEEE Internet of Things Magazine
DOI 10.1109/iotm.004.2100182
Open this paper in CoChat — free account

Abstract

Billions of IoT devices will be deployed in the near future, taking advantage of faster Internet speed and the possibility of orders of magnitude more endpoints brought by 5G/6G. With the growth of IoT devices, vast quantities of data that may contain users' private information will be generated. The high communication and storage costs, mixed with privacy concerns, will increasingly challenge the traditional eco-system of centralized over-the-cloud learning and processing for IoT platforms. Federated learning (FL) has emerged as the most promising alternative approach to this problem. In FL, training data-driven machine learning models is an act of collaboration between multiple clients without requiring the data to be brought to a central point, hence alleviating communication and storage costs and providing a great degree of user-level privacy. However, there are still some challenges existing in the real FL system implementation on IoT networks. In this article, we discuss the opportunities and challenges of FL in IoT platforms, as well as how it can enable diverse IoT applications. In particular, we identify and discuss seven critical challenges of FL in IoT platforms and highlight some recent promising approaches toward addressing them.

AI snapshot

Written from the paper’s full text.

TL;DR

This article surveys why federated learning (FL) suits the Internet of Things (IoT), describing key advantages and six application domains, then identifies seven critical challenges to real-world FL deployment on IoT devices and highlights recent approaches addressing them. It is a position/overview article rather than an empirical study.

Key findings

The paper identifies seven challenges as key barriers to enabling FL on potentially billions of IoT devices: (1) limited on-device resources, (2) limited network bandwidth, (3) intermittent connectivity and availability, (4) system heterogeneity, (5) temporal dynamics and continual learning, (6) trustworthiness, and (7) standardization and system development tools. It argues FL's distributed, collaborative, privacy-preserving nature yields three advantages for IoT: preserving user data privacy (raw data never leaves devices; only model updates are sent), improving model performance (devices collaboratively train a model beyond their own data), and flexible scalability (parallel use of constrained edge compute without centralizing data). It highlights example approaches per challenge, including FedMask for computation/communication efficiency, Mercury for sampling-based on-device training, FedBuff for buffered asynchronous aggregation, HeteroFL for heterogeneous local model architectures, FedGKT for reducing on-device training memory, DIoT for FL-based anomaly-detection intrusion detection, and FedIoT as a systems-level deployment framework for CPU/GPU IoT devices such as Raspberry Pi and NVIDIA Jetson Nano. It notes that current edge solutions (TensorFlow Lite, MNN, TVM) focus on inference rather than training, and that FL for IoT is not yet standardized.

Methods

Design: a conceptual overview/position article that explains FL advantages, surveys IoT applications, and enumerates challenges with opportunities; no experiments, datasets, or quantitative evaluation are reported. Applications discussed are Industry 4.0, healthcare, smart home, smart city, autonomous driving, and metaverse/virtual reality. Analysis: narrative synthesis of prior literature, citing representative systems and methods (e.g., FedMask, Mercury, FedBuff, HeteroFL, FedGKT, DIoT, FedIoT) as evidence for proposed directions.

Limitations

The paper states no formal limitations section; it frames its contribution as focusing on new challenges and articulating known challenges from new perspectives. Evident from the text: the work is a non-empirical survey with no experiments, benchmarks, or quantitative results, so its claims about which approaches are promising are not validated here; several proposed directions are explicitly described as envisioned rather than demonstrated (e.g., hardware-algorithm co-design, clustering to avoid catastrophic forgetting, lightweight secure aggregation).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/iotm.004.2100182

Cites

39 of 39 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 15 cited sources checked · 0 supported · 0 flagged

  • Interpretationreasoning
    § II. WHY FEDERATED LEARNING FOR IOT? · p. 2

    By attracting more devices to join the framework, FL enhances the scalability of IoT networks without adding an extra burden on a centralized server due to its distributed learning nature.

    “By attracting more devices to join the framework, FL enhances the scalability of IoT networks without adding an extra burden on a centralized server due to its distributed learning nature.”
    Quote in the paperSubstantial claim
  • Interpretationreasoning
    § II. WHY FEDERATED LEARNING FOR IOT? · p. 2

    FL is an effective mechanism to enhance the model performance that each individual device cannot achieve by itself.

    “Thus, FL is an effective mechanism to enhance the model performance that each individual device cannot achieve by itself.”
    Quote in the paperSubstantial claim
  • Interpretationreasoning
    § I. INTRODUCTION · p. 1

    Since raw data collected at each IoT device are not transmitted to others, FL provides an effective mechanism to protect user privacy particularly in the IoT domain where IoT sensors could directly capture data about users that contain privacy-sensitive personal information.

    “since the raw data collected at each IoT device are not transmitted to others, FL provides an effective mechanism to protect user privacy particularly in the IoT domain where IoT sensors could directly capture data about users that contain privacy-sensitive personal information.”
    Quote in the paperSubstantial claim
  • Interpretationreasoning
    § A. Limited On-device Resources · p. 3

    A significant challenge for processing neural network models on a resource constraint device is reducing the memory accesses and keeping the data on-chip to avoid costly reads and writes to external memory modules.

    “Therefore, a significant challenge for processing neural network models on a resource constraint device is reducing the memory accesses and keeping the data on-chip as to avoid costly reads and writes to the external memory modules.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Recent neural network architectures require accessing a vast amount of memory locations for storing not only model weights and parameters but also the intermediate results produced by the computations.
  • Premisereasoning
    § A. Limited On-device Resources · p. 3

    Despite the fact that current edge devices are increasingly powerful, training some deep learning models on-device is still time-consuming and inefficient.

    “Despite the fact that current edge devices are increasingly powerful, training some deep learning models on-device is still time-consuming and inefficient.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 28
  • May combine two claims. 22

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisereasoning
    § III. APPLICATIONS · p. 2

    FL has enabled many important IoT applications.

    “Benefited from the advantages mentioned above, FL has enabled many important IoT applications.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisereasoning
    § Federated Learning for Internet of Things: Applications, Challenges, and Opportunities · p. 1

    Federated Learning (FL) has emerged as the most promising alternative approach to centralized over-the-cloud learning and processing for IoT platforms.

    “Federated Learning (FL) has emerged as the most promising alternative approach to this problem.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § E. Autonomous Driving · p. 3

    With FL, each vehicle only needs to transmit a limited size of data to the cloud and can adapt to real-time local changes more sensitively.

    “With FL, each vehicle only needs to transmit a limited size of data to the cloud and can adapt to real-time local changes more sensitively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § II. WHY FEDERATED LEARNING FOR IOT? · p. 2

    Within the FL framework, there is no need for the expansive transmission of raw IoT-collected data, which also increases the scalability with regard to communication costs, especially for the low bandwidth IoT networks.

    “In addition, within the FL framework, there is no need for the expansive transmission of raw IoT-collected data, which also increases the scalability with regard to communication costs, especially for the low bandwidth IoT networks.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § B. Healthcare · p. 2

    Techniques such as FL are a requirement for investigators and researchers to develop state-of-the-art ML models over a fractured and highly regulated data landscape.

    “Therefore, techniques such as FL are a requirement for investigators and researchers to develop stateof-the-art ML models over a fractured and highly regulated data landscape.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisenone given
    § II. WHY FEDERATED LEARNING FOR IOT? · p. 1

    The raw data never leaves the devices during the federated training process, and only the updates of the model are sent to the central server, which minimizes the risk of personal data leakage.

    “The raw data never leaves the devices during the federated training process, and only the updates of the model are sent to the central server, which minimizes the risk of personal data leakage.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning
    § F. Metaverse and Virtual Reality · p. 3

    Federated learning is a promising solution to enable collaboration between edge and server for better global performance and also boost the security and privacy of the metaverse.

    “Federated learning is a promising solution to enable collaboration between edge and server for better global performance and also boost the security and privacy of the metaverse.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisenone given
    § II. WHY FEDERATED LEARNING FOR IOT? · p. 1

    In an ideal FL scenario, each IoT device in the system would learn nothing more than the information needed to play its role.

    “In an ideal FL scenario, each IoT device in the system would learn nothing more than the information needed to play its role.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given
    § Federated Learning for Internet of Things: Applications, Challenges, and Opportunities · p. 1

    In FL, training data-driven machine learning models is an act of collaboration between multiple clients without requiring the data to be brought to a central point, hence alleviating communication and storage costs and providing a great degree of user-level privacy.

    “In FL, training data-driven machine learning models is an act of collaboration between multiple clients without requiring the data to be brought to a central point, hence alleviating communication and storage costs and providing a great degree of user-level privacy.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § B. Healthcare · p. 2

    The ability to train machine learning models at scale across multiple medical institutions without pooling data is a critical technology to solve the problem of patient privacy and data protection.

    “The ability to train machine learning models at scale across multiple medical institutions without pooling data is a critical technology to solve the problem of patient privacy and data protection.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning
    § D. Smart City · p. 3

    Due to concerns of data privacy, smart infrastructures are moving to compute resources close to where data reside, which makes FL framework suitable for deployment.

    “Due to the concerns of data privacy, smart infrastructures are moving to compute resources close to where data reside, which makes FL framework suitable for deployment.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § C. Smart Home · p. 3

    FL has become a critical technology that is able to improve on-device speaker verification while reducing the risk of raw data leakage.

    “FL has thus become a critical technology that is able to improve the ondevice speaker verification while reducing the risk of raw data leakage.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning
    § A. Industry 4.0 · p. 2

    The federated learning framework will become an inspired solution to address the challenges of insufficient data from a single factory and privacy preservation in Industry 4.0.

    “The federated learning framework will become an inspired solution to address the above challenges.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning
    § D. Smart City · p. 3

    The FL-based smart grid system enables collaborative learning of power consumption patterns without leaking individual power traces and contributes to the establishment of an interconnected and intelligent energy exchange network in the city.

    “For example, the FL-based smart grid system enables collaborative learning of power consumption patterns without leaking individual power traces and contributes to the establishment of an interconnected and intelligent energy exchange network in the city.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning
    § I. INTRODUCTION · p. 1

    The distributed and collaborative nature of FL is a natural fit to the network edge where each IoT device at the edge is an individual client.

    “The distributed and collaborative nature of FL is a natural fit to the network edge where each IoT device at the edge is an individual client.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given
    § C. Intermittent Connectivity and Availability · p. 4

    Most of the FL studies are based on the synchronous update at the server, which implies that the server will not start the model aggregation until it receives the information sent from the slowest client.

    “Currently, most of the FL studies are based on the synchronous update at the server, which implies that the server will not start the model aggregation until it receives the information sent from the slowest client.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Most of the FL studies are based on the synchronous update at the server.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § E. Temporal Dynamics and Continual Learning · p. 5

    With FedGKT, IoT devices can transfer knowledge from many compact CNN models to a large CNN at a cloud server, reformulating federated learning as a group knowledge transfer training model for large-size model training on resource-constrained edge devices.

    “With FedGKT, IoT devices could transfer knowledge from many compact CNN models to a large CNN at a cloud server, which reformulates FL as a group knowledge transfer training model for the large-size model training on resource-constrained edge devices.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] He, 2020, Group Knowledge Transfer: Federated Learning of Large Cnns at the Edge

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation
    § F. Trustworthiness · p. 5

    An attacker can initiate a model inversion attack on the federated learning server to obtain the individual model of each participant and then recover personal training data by inverting these personal models.

    “The attacker could initiate the model inversion attack on the FL server first to achieve the individual model of each participant, and then recover the personal training data by inverting these personal models.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § E. Temporal Dynamics and Continual Learning · p. 5

    FedGKT [12] is a potential method to reduce the training memory footprint for efficient on-device learning.

    “As an example, FedGKT [12] is a potential method to reduce the training memory footprint for efficient on-device learning.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] He, 2020, Group Knowledge Transfer: Federated Learning of Large Cnns at the Edge

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning
    § C. Intermittent Connectivity and Availability · p. 4

    In real-world settings, due to the unbalanced communication abilities and training data distribution, the local training speed varies from device to device, and even some clients will be temporarily disconnected during the training phase, which makes the synchronous update nearly impossible.

    “However, in real-world settings, due to the unbalanced communication abilities and training data distribution, the local training speed varies from device to device, and even some clients will be temporarily disconnected during the training phase, which makes the synchronous update nearly impossible.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: In real-world settings, due to the unbalanced communication abilities and training data distribution, the local training speed varies from device to device, and even some clients will be temporarily disconnected during the training phase.

  • May combine two claims.

    Interpretationreasoning
    § C. Intermittent Connectivity and Availability · p. 4

    IoT devices with higher computing speed would be able to send in their local updates faster than IoT devices with slower computing speed, which inevitably leads to training bias.

    “IoT devices with higher computing speed would be able to send in their local updates faster than IoT devices with slower computing speed, which inevitably leads to training bias.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given
    § C. Intermittent Connectivity and Availability · p. 4

    In real-world settings, IoT devices are by nature heterogeneous with diverse computing speeds.

    “However, in real-world settings, IoT devices are by nature heterogeneous with diverse computing speeds.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: In real-world settings, IoT devices are by nature heterogeneous with diverse computing speeds.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § B. Limited Network Bandwidth · p. 4

    Chen et al.'s work formulates the bandwidth resource allocation and user selection problem during training federated learning models as an optimization problem whose goal is to minimize the training loss while meeting the delay and energy consumption requirements.

    “In addition, Chen et al.'s work [8] formulates the bandwidth resource allocation and user selection problem during training federated learning models as an optimization problem whose goal is to minimize the training loss while meeting the delay and energy consumption requirements.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [8] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § D. System Heterogeneity · p. 4

    In federated learning for IoT, the central server cannot examine the impact of data heterogeneity until training is complete.

    “However, the central server could not examine the impact of the data heterogeneity until the training is done.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: In IoT settings, data collected by different devices can differ in features and dimensions, and devices can have different temporal and spatial preferences for data collection, creating a discrepancy in local data structure among federated learning participants.

  • May combine two claims.

    Interpretationreasoning
    § D. System Heterogeneity · p. 4

    The diversity of hardware, software, and deep learning frameworks among clients in cross-device federated learning poses a significant challenge to system design and exacerbates asynchronous communication.

    “All the diversities have not only posed a significant challenge to system design but also exacerbated the asynchronous communication problem as mentioned above.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Within cross-device federated learning, clients have diverse system metrics in terms of both hardware and software.
    • Rests on an assumption: Clients may choose different deep learning frameworks such as TensorFlow, PyTorch, and Caffe to train local models, resulting in different model formats for aggregation.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § A. Limited On-device Resources · p. 4

    By applying FedMask, each device can learn a heterogeneous and structured sparse binary mask; based on the mask, it is able to generate a sparse model with reduced computation cost, memory footprint, and energy consumption.

    “By applying FedMask, each device can learn a heterogeneous and structured sparse binary mask; based on the mask, it is able to generate a sparse model with reduced computation cost, memory footprint, and energy consumption.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [6] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § C. Intermittent Connectivity and Availability · p. 4

    In FedBuff, the individual updates are not incorporated by the server as soon they arrive; instead, the server will keep receiving local updates in a secure buffer of size K, which is a tunable parameter, and then update the global model when the buffer is full.

    “Specifically, the individual updates are not incorporated by the server as soon they arrive. Instead, the server will keep receiving local updates in a secure buffer of size K, which is a tunable parameter, and then update the global model when the buffer is full.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Nguyen, 2021, Federated Learning with Buffered Asynchronous Aggregation

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § C. Intermittent Connectivity and Availability · p. 4

    In asynchronous FL scenarios, a client could join the active learning group even in the middle of the training progress, which endangers the convergence of the federated training.

    “In the asynchronous FL scenarios, a client could join the active learning group even in the middle of the training progress, which endangers the convergence of the federated training.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § C. Intermittent Connectivity and Availability · p. 4

    FedBuff is an asynchronous aggregation protocol that mitigates stragglers and enables secure aggregation jointly.

    “In [10], an asynchronous aggregation protocol known as FedBuff has been proposed to mitigate stragglers and enable secure aggregation jointly.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Nguyen, 2021, Federated Learning with Buffered Asynchronous Aggregation

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisenone given
    § C. Intermittent Connectivity and Availability · p. 4

    The intermittent connectivity of the IoT devices signifies an unstable network connection that drops the device out of the system in the middle of the training round.

    “Apart from the previous challenge of bandwidth limitations, the intermittent connectivity of the IoT devices signifies an unstable network connection that drops the device out of the system in the middle of the training round.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § B. Limited Network Bandwidth · p. 4

    Liu et al. proposed a client-edge-cloud hierarchical aggregation framework as a communication resource-efficient method to operate the federated learning in edge computing.

    “Liu et al. also proposed a client-edge-cloud hierarchical aggregation framework as a communication resource-efficient method to operate the federated learning in edge computing [9].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning
    § B. Limited Network Bandwidth · p. 4

    The limited network bandwidth not only makes the communication between clients and the server inefficient but also triggers the presence of straggler clients, which fail to share their local updates with the server during the communication round.

    “The limited network bandwidth not only makes the communication between clients and the server inefficient but also triggers the presence of straggler clients, which fail to share their local updates with the server during the communication round.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Most IoT devices communicate using wireless networks whose bandwidth is much smaller than wired network bandwidth in datacenters.

  • May combine two claims.

    Premisecitation
    § B. Limited Network Bandwidth · p. 4

    In the client-edge-cloud paradigm, each client is able to offload its data samples and learning tasks from its device to the edge in proximity for fast computation, which allows multiple edge servers to perform partial model aggregation.

    “Each client is able to offload its data samples and learning tasks from its device to the edge in proximity (e.g., edge gateway at home) for fast computation in the client-edge-cloud paradigm, which allows multiple edge servers to perform partial model aggregation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § A. Limited On-device Resources · p. 4

    FedMask is a joint computation and communication-efficient FL framework.

    “As an example, FedMask [6] is proposed as a joint computation and communication-efficient FL framework.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [6] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisereasoning
    § G. Standardization and System Development Tools · p. 6

    Federated learning for IoT has not been standardized and appropriately implemented.

    “As an emerging field, FL for IoT has not been standardized and appropriately implemented.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given
    § G. Standardization and System Development Tools · p. 6

    Current edge computing solutions such as TensorFlow Lite, MNN, and TVM focus on improving performance and efficiency of edge inference instead of training, and do not take federated learning settings into consideration.

    “It is worth mentioning that current edge computing solutions such as TensorFlow Lite, MNN, and TVM are focusing on improving the performance and efficiency of edge inference instead of training, much less taking FL setting into consideration, which is an under-explored area that would bring significant values to the federated learning and IoT communities.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given
    § V. CONCLUDING REMARKS · p. 6

    The distributed, collaborative, and privacy-preserving nature of federated learning makes it well suited for the IoT domain across a wide range of applications.

    “The distributed, collaborative, and privacy-preserving nature of federated learning makes it well suited for the IoT domain across a wide range of applications.”
    Quote in the paperNot a substantial claim
37 of 37

  • The paper relies on this without support.

    Billions of IoT devices will be deployed in the near future, taking advantage of faster Internet speed and 5G/6G.

    “Billions of IoT devices will be deployed in the near future, taking advantage of faster Internet speed and the possibility of orders of magnitude more endpoints brought by 5G/6G.”

  • The paper relies on this without support.

    The rapid advancement and expansion of IoT result in exponential growth of data being generated at the network edge.

    “The rapid advancement and expansion of the Internet of Things (IoT) result in exponential growth of data being generated at the network edge.”

  • The paper relies on this without support.

    Centralized approaches no longer fit the 5G/6G era due to extremely high communication and storage overhead for pooling data from millions or billions of IoT devices.

    “First, the centralized approaches no longer fit the 5G/6G era due to the extremely high communication and storage overhead (e.g., high-frequency data from high-volume time-series sensors such as video cameras or Lidar sensors) for pooling data from millions or billions of IoT devices.”

  • The paper relies on this without support.

    Data being collected is increasingly viewed as threatening user privacy.

    “Second, the data being collected is increasingly viewed as threatening user privacy.”

  • The paper relies on this without support.

    With cloud-based centralized approaches, user data could be shared between or even sold to various companies, violating privacy rights and negatively affecting data security, further driving public distrust with data-driven applications.

    “With the cloud-based centralized approaches, user data could be shared between or even sold to various companies, violating privacy rights and negatively affecting data security, further driving public distrust with datadriven applications.”

  • The paper relies on this without support.

    Due to device constraints, a single IoT device may not have sufficient data to learn a high-quality model by itself.

    “Due to device constraints, a single IoT device may not have sufficient data to learn a high-quality model by itself.”

  • The paper relies on this without support.

    As edge device hardware capability is increasing, the data size of each individual becomes huge, and centralizing all data to the server either wastes the computing resource at the edge or brings pressure for wireless communication network, which become an obstacle for the network scalability.

    “As edge device hardware capability is increasing, the data size of each individual becomes huge, and centralizing all data to the server either wastes the computing resource at the edge or brings pressure for wireless communication network, which become an obstacle for the network scalability.”

  • The paper relies on this without support.

    The amount of data generated from a single factory may not be sufficient enough for training a reliable model comprehensively.

    “First, the amount of data generated from a single factory may not be sufficient enough for training a reliable model comprehensively.”

  • The paper relies on this without support.

    The data collected by the industrial IoT devices is highly related to the commercial value, which makes privacy-preserving important.

    “Second, the data collected by the industrial IoT devices is highly related to the commercial value, which makes privacy-preserving important.”

  • The paper relies on this without support.

    Personal healthcare data is most sensitive to users' privacy and highly restricted by government laws and regulations for any kind of data sharing.

    “Compared to the other types of data, personal healthcare data is most sensitive to users' privacy and highly restricted by government laws and regulations for any kind of data sharing.”

  • The paper relies on this without support.

    Existing machine learning models, especially deep neural networks, are known to be computation-intensive, which presents strict requirements on hardware and may result in low training efficiency on edge devices.

    “Existing machine learning models, especially deep neural networks, are known to be computation-intensive, which presents strict requirements on hardware and may result in low training efficiency on edge devices.”

  • The paper relies on this without support.

    Edge devices have limited resources not only in terms of computation but also in terms of memory for storage and data access.

    “Edge devices have limited resources not only in terms of computation but also in terms of memory for storage and data access.”

  • The paper relies on this without support.

    Recent neural network architectures require accessing a vast amount of memory locations for storing not only model weights and parameters but also the intermediate results produced by the computations.

    “Recent neural network architectures require accessing a vast amount of memory locations for storing not only model weights and parameters but also the intermediate results produced by the computations.”

  • The paper relies on this without support.

    Edge devices with embedded processors have a limited energy budget, which further imposes restrictions on the hardware performance.

    “Finally, in contrast to servers with CPUs and GPUs that can use a substantial amount of power, edge devices with embedded processors have a limited energy budget, which further imposes restrictions on the hardware performance.”

  • The paper relies on this without support.

    Most IoT devices communicate using wireless networks whose bandwidth is much smaller than wired network bandwidth in datacenters.

    “Currently, most IoT devices communicate using wireless networks whose bandwidth is much smaller than wired network bandwidth in datacenters.”

  • The paper relies on this without support.

    Under a smart home scenario, the sum of the overall networking bandwidth is constant for the whole IoT system, no matter how many devices are connected.

    “For example, under a smart home scenario, the sum of the overall networking bandwidth is constant for the whole IoT system, no matter how many devices are connected.”

  • The paper relies on this without support.

    Most of the FL studies are based on the synchronous update at the server.

    “Currently, most of the FL studies are based on the synchronous update at the server, which implies that the server will not start the model aggregation until it receives the information sent from the slowest client.”

  • The paper relies on this without support.

    In real-world settings, due to the unbalanced communication abilities and training data distribution, the local training speed varies from device to device, and even some clients will be temporarily disconnected during the training phase.

    “However, in real-world settings, due to the unbalanced communication abilities and training data distribution, the local training speed varies from device to device, and even some clients will be temporarily disconnected during the training phase, which makes the synchronous update nearly impossible.”

  • The paper relies on this without support.

    In real-world settings, IoT devices are by nature heterogeneous with diverse computing speeds.

    “However, in real-world settings, IoT devices are by nature heterogeneous with diverse computing speeds.”

  • The paper relies on this without support.

    Within cross-device federated learning, clients have diverse system metrics in terms of both hardware and software.

    “Within cross-device settings, clients under the FL framework have diverse system metrics in terms of both hardware and software.”

  • The paper relies on this without support.

    Clients may choose different deep learning frameworks such as TensorFlow, PyTorch, and Caffe to train local models, resulting in different model formats for aggregation.

    “Clients may choose different deep learning frameworks such as TensorFlow, PyTorch, and Caffe to train the local models, resulting in different model formats for aggregation.”

  • The paper relies on this without support.

    In IoT settings, data collected by different devices can differ in features and dimensions, and devices can have different temporal and spatial preferences for data collection, creating a discrepancy in local data structure among federated learning participants.

    “Moreover, in IoT settings, the data collected by different devices can be very different in terms of the feature and dimensions, and various types of devices can also have different temporal and spatial preferences for their data collection, which may create a discrepancy in the local data structure among all the participants under FL framework.”

  • The paper relies on this without support.

    IoT devices are attractive targets for adversaries seeking to launch attacks such as phishing, identity theft, and distributed denial of service (DDoS).

    “In practical deployment, IoT devices are attractive targets for adversaries seeking to launch attacks such as phishing, identity theft, and distributed denial of service (DDoS).”

  • The paper relies on this without support.

    Many IoT devices lack the requisite computation resources to install security patches.

    “Even though these attacks could be easily defended by installing security patches, many IoT devices lack the requisite computation resources to do so.”

  • The paper relies on this without support.

    Within a cross-device system setting, it is difficult to identify whether a coming participant is malicious before it joins the system.

    “Moreover, within a cross-device system setting, it is difficult to identify whether the coming participant is malicious or not before it joins the system.”

  • The paper relies on this without support.

    IoT sensing devices continuously collect new data, which will be used to update the model for lifelong or continual learning.

    “IoT sensing devices will, by their very nature, continuously collect new data, which will be used to update the model for lifelong or continual learning.”

  • The paper relies on this without support.

    Most IoT devices are memory-limited, and their memory resources are not sufficient to handle both model inference and training.

    “Specifically, as most of the IoT devices are memory-limited, their memory resources are not sufficient enough to handle both model inference and training.”

  • The paper relies on this without support.

    Lack of sufficient memory to keep past collected data may exacerbate catastrophic forgetting, which is a critical problem in continual learning.

    “Furthermore, the lack of sufficient memory to keep past collected data may exacerbate catastrophic forgetting, which is one of the most critical problems in continual learning.”
28 of 28

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?

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. 10.1561/9781680837896 · DOI
  2. Diao, 2020, Heterofl: Computation and Communication Efficient Federated Learning for Heterogeneous Clients
  3. He, 2020, Group Knowledge Transfer: Federated Learning of Large Cnns at the Edge, 33
  4. Geiping, 2020, Inverting Gradients-How Easy Is It to Break Privacy in Federated Learning?, 33

Cited by

Sorted by citations

Related

Authors
Zhang, T., Gao, L., He, C., Zhang, M., Krishnamachari, B., & Avestimehr, A. S.
Affiliations
  • University of Southern California (US)
  • Michigan State University (US)
ORCID
Venue
IEEE Internet of Things Magazine
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2022-03
Volume
5
Issue
1
Pages
24-29
ISSN
  • 2576-3180 (ISSN-L)
  • 2576-3199
DOI
10.1109/iotm.004.2100182
OpenAlex
W3213321731
Semantic Scholar
32cf329348be58d60eb051d2071f02945fe33bbc
Cited by
412
References
39
Citations by year
2026: 73 · 2025: 124 · 2024: 132 · 2023: 68 · 2022: 14 · 2021: 1
FWCI
27.82
Type
journal article
Language
en
Open access
closed
Licence
Locations
IEEE Internet of Things Magazine (publishedVersion)
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 › Information Systems › Internet of Things and AI
Keywords
federated learning, Internet of Things, privacy-preserving machine learning, IoT platforms, 5G/6G networks, distributed learning

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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