PaperlayerView exact revision on arXiv

Source-linked AI summary

Explainable AI over the Internet of Things (IoT): Overview, State-of-the-Art and Future Directions

Senthil Kumar Jagatheesaperumal, Quoc-Viet Pham, Rukhsana Ruby, Zhaohui Yang, Chunmei Xu, Zhaoyang Zhang

arXiv:2211.01036v2cs.AIcs.HCcs.LG

TL;DR

Heterogeneous IoT systems make interoperability difficult, while the paper surveys XAI characteristics, applications, and adaptive IoT solutions. It highlights edge XAI's interpretability, quick response, and accessibility, and presents it as a basis for large-scale IoT integration.

  • Problem

    Heterogeneous IoT systems exchange information across diverse devices, formats, and data models, making commonly accepted interoperability specifications difficult.

    Source: 2) Interoperability challenges and… · p. 17

  • Method

    The paper surveys XAI characteristics and IoT use cases, including adaptive solutions based on 5G services, cloud services, and big-data management.

    Source: §VI · p. 23

  • Results

    The survey highlights edge XAI structures for healthcare and smart-city IoT applications for their interpretability, quick response time, and accessibility.

    Source: C. Summary · p. 22

  • Takeaways & Limitations

    The authors present edge XAI structures as a means for long-term, massive integration of IoT devices operating on large volumes of data.

    Source: C. Summary · p. 22

    Interoperability is constrained by diverse systems, devices, sensors, equipment, data formats, and models, making shared specifications difficult to establish.

    Source: 2) Interoperability challenges and… · p. 17

Abstract

from arXiv · show

Explainable Artificial Intelligence (XAI) is transforming the field of Artificial Intelligence (AI) by enhancing the trust of end-users in machines. As the number of connected devices keeps on growing, the Internet of Things (IoT) market needs to be trustworthy for the end-users. However, existing literature still lacks a systematic and comprehensive survey work on the use of XAI for IoT. To bridge this lacking, in this paper, we address the XAI frameworks with a focus on their characteristics and support for IoT. We illustrate the widely-used XAI services for IoT applications, such as security enhancement, Internet of Medical Things (IoMT), Industrial IoT (IIoT), and Internet of City Things (IoCT). We also suggest the implementation choice of XAI models over IoT systems in these applications with appropriate examples and summarize the key inferences for future works. Moreover, we present the cutting-edge development in edge XAI structures and the support of sixth-generation (6G) communication services for IoT applications, along with key inferences. In a nutshell, this paper constitutes the first holistic compilation on the development of XAI-based frameworks tailored for the demands of future IoT use cases.

I. INTRODUCTION

The introduction argues that IoT needs trustworthy, interpretable AI while XAI support remains constrained by device resources and challenges in explanation, privacy, and deployment. It positions XAI as a way to make model decisions more transparent and to support trust in IoT applications.

I. INTRODUCTION

  • XAI has applications across healthcare, finance, security, military, and legal sectors, with examples including defense and cloud services.The introduction presents these as areas attracting industry and academic interest.Source: §I · p. 1
  • In IoT, interpretable ML models aim to explain black-box decisions, but high-risk applications must account for random new samples.The introduction also describes LIME-based models as learning domain-invariant features for trustworthy processing and faithful explanations.Source: §I · p. 1

B. Role of XAI in IoT

  • The paper frames XAI as a route to trustworthy IoT systems and reviews its use across IoT roles, including sensor-data analysis and supervision of AI elements.It notes that IoT activity recognition commonly relies on ML and DL techniques.Source: §I · p. 2, B. Role of XAI in IoT · p. 2

A. Motivation

B. Role of XAI in IoT

  • XAI can make AI decisions more understandable to people, supporting interpretation of models and transparent monitoring of patient activity.The paper contrasts this with conventional AI, which does not explain its decisions to humans.Source: B. Role of XAI in IoT · p. 2

C. Comparison and Our Contributions

The authors identify a gap in dedicated, comprehensive surveys of XAI across IoT services and applications. They propose a broad review spanning IoT and XAI foundations, application domains, dependable architectures, and future challenges.

C. Comparison and Our Contributions

  • The authors identify no comprehensive, dedicated survey of XAI across IoT services and applications, and provide a review spanning security, healthcare, industry, and smart cities.They also summarize observations from applications and outline research challenges and future directions.Source: C. Comparison and Our Contributions · p. 3
  • The survey reviews background on IoT, XAI, and their integration, and presents the need for XAI in IoT and challenges reported in recent literature.Table I is described as covering existing XAI-related surveys and the authors’ new contributions, with existing survey contributions and limitations summarized in the surrounding text.Source: C. Comparison and Our Contributions · p. 3, Table I · p. 4
  • The application review covers XAI in security, healthcare, industry, and smart cities, including its role and lessons for trustworthy IoT services.The authors also describe their work as a detailed discussion of developments for dependable IoT services using XAI models.Source: C. Comparison and Our Contributions · p. 3

D. Paper Structure and Organization

  • The paper is organized to introduce XAI and IoT fundamentals, survey applications and open challenges, summarize findings, discuss future directions, and conclude.The preliminary section introduces XAI and IoT concepts and their key principles.Source: D. Paper Structure and Organization · p. 3, §II · pp. 3, 5

A. Explainable AI

XAI aims to make AI decisions and learning processes understandable to users, addressing the opacity of complex models. Its value includes supporting model improvement, communication, and avoidance of unforeseen consequences, while accuracy and explainability can trade off.

A. Explainable AI

  • XAI makes AI decisions and learning processes more transparent to end-users than conventional systems, which present learning as an opaque black box.XAI relates model outputs to inputs and feature contributions to explain how decisions are reached.Source: A. Explainable AI · p. 5
  • The paper distinguishes interpretability, transparency, and trustworthiness as related XAI concepts concerning understandable decisions, inspectable model workings, and mechanisms supporting user trust.White-box models are described as exposing behavior, prediction processes, and influencing factors.Source: A. Explainable AI · p. 5
  • Deep-learning models offer better accuracy with less explainability, while decision-tree, random-forest, and ensemble models offer greater explainability with lower performance accuracy.The paper identifies trade-offs among model complexity, explanatory capability, and system performance.Source: A. Explainable AI · p. 5
  • Complex deep-learning and ensemble techniques can improve accuracy but make input-to-output relationships harder to understand; these are described as black boxes.The paper notes that tolerance for this trade-off depends on the decision context, such as movie recommendations versus credit-card rejection.Source: A. Explainable AI · p. 6
  • Explainability can help refine analysis, explain decisions to non-practitioners, and avoid negative or unforeseen model consequences.The paper links these benefits to greater confidence in model development and deployment.Source: A. Explainable AI · p. 6

1) XAI Architectures for Transparency in AI:

The paper groups XAI architectures by how they represent, communicate, and explain decisions, then surveys frameworks that provide model explanations through different approaches. These choices expose trade-offs in accuracy, explainability, interaction, and implementation.

1) XAI Architectures for Transparency in AI:

  • Probabilistic XAI represents uncertainty through distributions and random variables, producing ranges of potential outcomes rather than a single best guess.The explanations can help users understand forecasting risks and fine-tune systems.Source: 1) XAI Architectures for Transparency… · p. 7
  • Multi-component XAI uses interactive, dialogue-based explanations when a single explanation cannot capture system operations.The paper says this approach can enhance social acceptance and trust.Source: 1) XAI Architectures for Transparency… · p. 7
  • Visual analytics represents decision-making interactively and iteratively, helping users assess model inputs, outputs, and the stages leading to an action.Model-based interpretation is also presented as supporting fairness, transparency, accountability, and debugging.Source: 1) XAI Architectures for Transparency… · pp. 6–7
  • Deep learning can provide robust explanations, but their accuracy depends on data quality and size, feature extraction, and model configuration.The paper also contrasts higher accuracy but lower explainability in deep learning with more explainable decision-tree, random-forest, and ensemble approaches.Source: 1) XAI Architectures for Transparency… · p. 7

2) XAI Frameworks for Transparency in AI:

  • The surveyed frameworks offer distinct explanation modes: SHAP attributes feature contributions, LIME fits a simple local model, and ELI5 explains predictions and model characteristics.WIT provides a visual interface for dataset and output analysis across model creation and evaluation; AIX360 and Skater provide toolkits for explaining model decisions and local or global predictions.Source: 2) XAI Frameworks for Transparency in AI · pp. 7–8

B. Internet of Things (IoT)

IoT systems generate large and varied data, and AI can support prediction and automation across connected applications. The paper positions XAI as a means to make IoT analytics and decisions more trustworthy and understandable, with examples across application domains.

B. Internet of Things (IoT)

  • IoT devices generate large volumes and varieties of data that conventional processing can struggle to manage, while AI can support accurate, rapid prediction and insight.The paper describes applications spanning healthcare, industry, homes, and other smart systems.Source: 1) Role of AI in IoT · p. 8, B. Internet of Things (IoT) · p. 8

1) Role of AI in IoT:

  • AI integrated with IoT can help manage industrial downtime, automate product inspection, identify defects, and improve operational efficiency.The paper also notes AI-enabled wireless architectures for processing and transferring data to server and cloud services for remote monitoring and control.Source: 1) Role of AI in IoT · p. 8, 2) Trustworthiness of IoT Systems · p. 8

2) Trustworthiness of IoT Systems:

  • XAI can make IoT operations and decisions more transparent, addressing privacy and security concerns associated with heterogeneous connected technologies.Prior work surveyed IoT trust requirements and proposed adaptive trust-management protocols.Source: 2) Trustworthiness of IoT Systems · p. 9

III. APPLICATION DOMAINS OF XAI OVER IOT USE CASES

  • The paper surveys XAI use across IoT applications including security, healthcare, robotics, industrial IoT, and commercial IoT.It frames XAI as supporting trustworthy collection, processing, analysis, and use of data for autonomous tasks.Source: §III · p. 9, A. Enabling XAI in IoT systems: An… · p. 9

A. Enabling XAI in IoT systems: An Overview

  • At the center of IoT systems, XAI can support predictive analytics, anomaly alerts, and decisions; examples include smart-machine maintenance and smart-city services.Other examples include interpretable fuzzy-rule irrigation and a smart-kitchen simulator whose generated explanations reveal human-AI interactions used to detect depression from meal intake.Source: A. Enabling XAI in IoT systems: An… · p. 9

B. Security Enhancement in IoT Devices

IoT devices face security challenges that include weak credentials, limited resources, physical access, and long operational lifetimes. The paper reviews AI-based security capabilities and XAI approaches for detecting and interpreting cyber threats.

B. Security Enhancement in IoT Devices

  • IoT security is constrained by weak or static credentials, limited device resources, physical attacks, and long device lifecycles.The paper notes that many devices lack built-in security support and cannot run security solutions designed for general-purpose systems.Source: B. Security Enhancement in IoT Devices · p. 9

1) Security Enhancement in IoT Devices with AI:

  • AI can process large IoT data volumes, learn to recognize attacks, identify unknown threats, and detect threats quickly, but it can also be exploited for criminal activity.AI can automate device authentication and strengthen passwords and biometric security features.Source: 1) Security Enhancement in IoT Devices… · pp. 9–10

2) XAI for Security Enhancement in IoT Devices:

  • An LSTM-based deep-learning detector for IoT cyberattacks reportedly achieved around 99% detection accuracy and can help explain its decisions.A separate convolutional LSTM encoder-decoder predicts anomalous messages between surveillance-broadcast and air-traffic-control systems and explains operational information to pilots [83].Source: 2) XAI for Security Enhancement in IoT… · p. 10
  • XAI security work also includes gradient-based classifier explanations, domain-knowledge infusion, feedback mechanisms for intrusion detection, and cloud-edge security assessment.The cited work reports interactive visualizations for diagnosing NSL-KDD99 intrusion-detection decisions, while domain knowledge is associated with better explainability and quicker decision-making [85].Source: 2) XAI for Security Enhancement in IoT… · p. 10

3) Lessons Learned:

The section connects trustworthy IoT security to transparent XAI decisions, while identifying deployment and training trade-offs. It also describes IoT-supported remote healthcare and future security applications.

3) Lessons Learned:

  • XAI can make IoT attacks transparent and visualize their impact, supporting countermeasures and secure data transmission, including through B5G/6G-supported systems.The section also links understandable decisions to more robust IoT frameworks and anticipation of security threats.Source: 3) Lessons Learned · p. 10
  • IoT security requires reliable infrastructure across data collection, exchange, cloud communication, storage, and analytics, with risks persisting whether data are in motion, at rest, or being processed.The section describes XAI as a way to establish more reliable and transparent cyber-defense decisions for mitigating risks.Source: 3) Lessons Learned · pp. 10–11
  • XAI security models are trained and evaluated across datasets and feature sets, but training overhead and robustness must be balanced against time complexity, resources, and accuracy.The section identifies generalisability evaluation as a common resource for strengthening AI-driven IoT security.Source: 3) Lessons Learned · p. 11
  • Operational deployment requires identifying IoT data sources and an appropriate accumulation space, then training on diversified data from interconnected devices to classify threats accurately.Source: 3) Lessons Learned · p. 11

4) Future Works:

  • Rule-based explanations of attack behavior and impact can help security personnel distinguish malicious nodes from regular traffic and plan against intrusions.The section proposes extending XAI security capabilities through transparent, explainable solutions.Source: 4) Future Works · p. 11

1) Healthcare Support Using IoT-based Systems:

  • IoT telehealth can support remote monitoring and doctor-patient interaction, while wireless PAN-connected medical devices feed data to XAI decisions that assist patients and healthcare professionals.Healthcare data may be generated in large volumes and need rapid processing; the described sensor networks require access points for broad-area communication.Source: 1) Healthcare Support Using IoT-based… · p. 11

2) AI-driven IoT Systems for Healthcare Support:

The healthcare discussion moves from AI analysis of medical and sensor data to XAI-based IoMT diagnosis and human-centric explanations. It highlights both the promise of transparent decisions and the need to manage analytics choices.

2) AI-driven IoT Systems for Healthcare Support:

  • AI and ML/DL approaches are used to analyze high-dimensional medical data, support health predictions and decisions, and manage healthcare information.Source: 2) AI-driven IoT Systems for… · p. 12
  • EEG analysis faces noisy signals and difficult interpretation, while deep-learning methods are applied from noise segmentation through signal classification.The passage says interpreting brain signals requires extreme precision.Source: 2) AI-driven IoT Systems for… · p. 12

3) XAI for IoT-based Healthcare Support Systems:

  • IoMT-driven XAI models are presented for disease detection and prediction, with transparent procedures and reasoning intended to address moral and legal implications of diagnosis.Source: 3) XAI for IoT-based Healthcare… · p. 12

4) Lessons Learned:

  • The healthcare XAI literature reports transparent models in medical practice and smart wearables, while intra-variability remains despite improved ML-based diagnosis and deep-learning signal classification.The section links these applications to greater user confidence in healthcare AI.Source: 4) Lessons Learned · p. 12
  • Human-centric AI is proposed for trustworthy IoT devices that provide interpretable decisions and can help users detect unreliable predictions, including in healthcare services.The passage gives patient depression tracking and explanatory alerts to family, friends, and doctors as an example.Source: 4) Lessons Learned · p. 13

5) Future Works:

The future-works discussion describes wearable healthcare monitoring and data-governance needs alongside IIoT manufacturing and communication approaches. Its examples span posture tracking, condition-based maintenance, analytics, and industrial networking.

5) Future Works:

  • IoMT wearables and sensors can monitor posture in real time and use XAI-guided alerts to help users adjust habits, while healthcare data integrity and reliability remain significant challenges.The section calls for a big-data governance environment because these data are associated with patients' lives.Source: 5) Future Works · p. 13

D. Industrial IoT

  • AI-enabled robots combine sensors and cameras with AI decision-making, contrasting with earlier industrial robots described as performing programmed tasks without intelligent decisions.Source: D. Industrial IoT · p. 13

1) The IIoT and Smart Manufacturing:

  • IIoT smart manufacturing uses machine data for analytics and bottleneck management; cited work proposes ARIMA prediction for machine behavior during active periods.A three-stage ensemble approach for condition-based maintenance trains the ensemble, detects imbalanced IIoT data, and creates new ensembles; its experimental analysis reports high accuracy in detecting all data drifts.Source: 1) The IIoT and Smart Manufacturing · pp. 13–14
  • Text mining is described as supporting industrial operations, operator skills, and business opportunities, while MapReduce can minimize data transfer across nodes when managing large IIoT datasets.Source: 1) The IIoT and Smart Manufacturing · p. 14

2) AI-driven IIoT:

  • For AI-driven IIoT, 5G, AI, and Big Data are linked to faster industrial-data analysis, while network slicing is proposed to meet IIoT devices' QoS needs for reliability, delay, and bandwidth.The section also describes sensor fusion for e-maintenance and Tactile Internet with 5G for high-performance, low-latency, ultra-reliable industrial communication.Source: 2) AI-driven IIoT · p. 14

3) XAI for the IIoT:

The section surveys security and trust concerns in IIoT, then outlines broader lessons, future communication needs, and IoCT contexts. It connects smart-city infrastructure and connected homes with XAI attention.

3) XAI for the IIoT:

  • Industry 4.0 cybersecurity requires collaboration between IT experts and manufacturing stakeholders to strengthen privacy and security across the digital ecosystem.The section also describes trust mechanisms and secure data, production, and connectivity systems for industrial environments.Source: 3) XAI for the IIoT · p. 14

4) Lessons Learned:

  • Industry 4.0 lessons emphasize vertical and horizontal data integration, which can require upgrading or replacing equipment, networks, and processes.The section situates this need within broader industrial digitization and evolving smart-industry technologies.Source: 4) Lessons Learned · p. 15

5) Future Works:

  • Although 5G supports flexible manufacturing and automation, the transition to 6G and beyond still faces unresolved technical challenges.The section describes centralized computing and wireless services as part of the evolving industrial IoT landscape.Source: 5) Future Works · p. 15

E. Internet of City Things

  • IoCT includes connected homes and buildings, where Internet-connected devices support remote appliance monitoring and management.The section identifies IoHT as an integral component of IoCT and points to sectors seeking XAI attention.Source: E. Internet of City Things · p. 15, Figure 6 · p. 15

1) Smart City IoT Products:

The section describes smart-city products and services, then discusses AI applications and the need for explanations as connected systems grow. Its lessons connect trustworthy data use with communication and energy-management requirements.

1) Smart City IoT Products:

  • An IoT- and augmented-reality-based disaster-management system provides rescue guidance to residents in smart buildings during emergencies.The work includes interoperable IoT devices and a smart-city testbed.Source: 1) Smart City IoT Products · pp. 15–16

2) AI driven Smart City IoT Products:

  • As connected IoCT devices increase, conventional AI frameworks need additional features to explain their decisions in smart-city applications.The section describes AI use in smart-building parameters and automated traffic classification.Source: 2) AI driven Smart City IoT Products · p. 16

3) XAI for Smart City IoT Products:

4) Lessons Learned:

  • Smart-city applications require high-speed communication and trustworthy data transfer, while XAI can help identify energy use and support energy-saving awareness.Long periods of fault-free operation in automation are also identified as a challenge for energy-efficient solutions.Source: 4) Lessons Learned · pp. 16–17

5) Future Works:

The paper identifies practical implementation issues for smart-city XAI and interoperability, then surveys edge-XAI developments and distributed reinforcement learning. It emphasizes resource constraints and the need to optimize edge deployments.

5) Future Works:

  • Smart-city XAI implementation still faces practical issues involving hardware design, security, deployment-site control, and resource management.The section identifies infrastructure investment opportunities in buildings and construction, energy, water and waste management, and enabling technologies.Source: 5) Future Works · p. 17

F. Summary

  • XAI and IoT can support trustworthy services and transparent views of device decisions, but emerging research challenges remain.The summary specifically frames interoperability among XAI-based IoT applications as an area of advantages, challenges, and limitations.Source: F. Summary · p. 17

1) Advantages of XAI based IoT Systems:

  • XAI can expose the decision-making stages of AI systems that analyze data generated by IoT devices.The passage contrasts this with AI models that do not exhibit those stages to end-users.Source: 1) Advantages of XAI based IoT Systems · p. 17

2) Interoperability challenges and limitations of XAI based IoT Systems:

  • Interoperability is difficult because diverse systems and devices exchange time- and location-dependent information in varied formats, languages, and data models.The passage notes that reaching commonly accepted specifications is difficult across changing designers, manufacturers, and application domains.Source: 2) Interoperability challenges and… · p. 17

IV. XAI: CUTTING EDGE DEVELOPMENTS FOR DEPENDABLE IOT SYSTEMS

  • Edge deployment is motivated by cloud scalability and bandwidth constraints, while edge XAI is presented as a way to better understand and validate AI decision mechanisms.Table II is identified as a summary of cutting-edge developments for dependable IoT systems.Source: A. Edge XAI Structures · p. 17, §IV · p. 17, Table II · p. 18

A. Edge XAI Structures

  • Distributed reinforcement learning can optimize training time and increase throughput and concurrency for large datasets, but its use in edge XAI requires resource and computation-cost optimization.The section says joint optimization is challenging and can enhance edge-XAI performance for IoT applications.Source: 1) Distributed Reinforcement Learning · p. 19

1) Distributed Reinforcement Learning:

  • Distributed reinforcement learning is discussed across applications including wireless networks, urban driving, and UAV and mobile-edge-computing systems.The cited work also considers power control, real-world decision-making, and continuous control.Source: 1) Distributed Reinforcement Learning · p. 19

2) Distributed Supervised Learning:

The section reviews distributed supervised learning, federated learning, and blockchain-based distributed learning for IoT contexts. It describes privacy and resource benefits alongside requirements for trustworthy, resource-constrained edge execution.

2) Distributed Supervised Learning:

  • Distributed supervised learning supports parallel processing and training of large IoT datasets while interpreting multiple distributed models.The section also discusses cognitive features and cooperation among learning tasks in distributed architectures.Source: 2) Distributed Supervised Learning · p. 19

3) Federated Learning:

  • Federated learning trains across devices through local training and server-side model combination without direct server access to training data.Its privacy-preserving accuracy depends on communication, edge infrastructure, and device-specific parameters.Source: 3) Federated Learning · p. 19

4) Blockchain Based Distributed Learning:

  • Blockchain-based distributed learning can remove trusted third parties, while edge healthcare examples combine blockchain smart contracts with federated learning for authentication and trust management.The described healthcare framework also uses full encryption on collected data and additive encryption at edge nodes.Source: 4) Blockchain Based Distributed Learning · pp. 19–20
  • Edge learning frameworks need to be secure and trustworthy while executing AI algorithms on resource-constrained devices.The section calls for innovative measures to implement trustworthy XAI structures efficiently.Source: 4) Blockchain Based Distributed Learning · p. 20

B. Potential to Meet 6G Requirements for IoT

The section surveys 6G capabilities relevant to XAI-driven IoT and summarizes both edge-XAI opportunities and gaps in integrating 6G services. It covers mURLLC, scalable architecture, massive access, and green communication, alongside future research needs.

1) Massive ultra-reliable, low latency communications (mURLLC):

  • 6G mURLLC targets latency-sensitive IoT, with a target latency around one microsecond and trade-offs among throughput, reliability, and latency.The cited work reports that 6G-enabled edge intelligence improves network mobility and communication latency.Source: 1) Massive ultra-reliable, low latency… · p. 20

2) Scalable architecture:

  • Scalable 6G IoT architecture must support parallel computing across multiple edge devices as the number of integrated devices grows.The survey also links scalable architecture and FL-based 6G services to reduced security threats and distributed computation resources.Source: 2) Scalable architecture · pp. 20–21

3) Massive access:

  • 6G massive access uses multi-antenna base stations to serve many single-antenna IoT devices and address bandwidth efficiency, latency, and heterogeneous traffic demands.The surveyed approaches include grant-free access and low-energy multiple access techniques.Source: 3) Massive access · p. 21

4) Green communication:

  • Green communication combines renewable energy and power-efficient network operation, making it a potential fit for IoT devices.The survey describes approaches including a base-station wakeup strategy and learning-based resource allocation [199, 200].Source: 4) Green communication · p. 21

C. Summary

  • Edge XAI can offer interpretable characteristics, quick response, and better accessibility in healthcare and smart-city IoT, while large-scale integration and 6G adoption still face infrastructure gaps.The survey identifies RL, FL, and blockchain-based services as support for XAI frameworks; it notes that 6G integration is limited by a lack of enabling services.Source: C. Summary · p. 22

V. FUTURE RESEARCH DIRECTIONS AND CONSIDERATION OF XAI FOR IOT

  • Future XAI-over-IoT implementation needs work on THz and semantic communication, energy efficiency, synchronization, signal-processing models, and low-cost, low-power hardware.The survey names these as research areas for practical IoT systems.Source: §V · p. 22

A. THz Communication

The future directions section identifies communication, energy, hardware, distributed-system, security, and accountability challenges for XAI-enabled IoT. The survey concludes with an overview of XAI solutions and research gaps across IoT systems and applications.

A. THz Communication

  • THz communication offers potential capacity and sensing gains for IoT, but channel uncertainty, blockage, molecular absorption, and network-design challenges remain.AI analysis of IoT data is described as a way to help improve THz communication performance and QoS.Source: A. THz Communication · p. 22

B. Semantic Communication

  • Semantic communication must balance transparent XAI decisions with efficient transmission in IoT, while maintaining trustworthiness.The section identifies DL-based semantic communication as a technique for improving transmission efficiency [205].Source: B. Semantic Communication · p. 22

C. High Energy Efficiency Issues

  • Energy-efficient XAI remains challenging because IoT deployments are distributed and energy techniques vary across scenarios and frameworks.The section calls for flexible, low-complexity XAI models to support trustworthiness with energy efficiency.Source: C. High Energy Efficiency Issues · p. 22

D. Hardware Design for XAI Devices over IoT

  • Local processing of IoT data and information exchange among XAI-enabled devices make energy-efficient hardware and renewed synchronization protocols important design needs.Neural-network size, layer count, and activation functions affect operations and computational complexity.Source: D. Hardware Design for XAI Devices… · pp. 22–23

E. Distributed XAI

  • Distributed XAI must combine decisions from multiple models while preserving transparent, faithful, and trustworthy interpretation of stage-wise decisions.The section calls for evidence and alternative approaches to distributing IoT data to models and accumulating their decisions.Source: E. Distributed XAI · p. 23

F. XAI Security and Privacy

  • XAI security research needs to address privacy, confidentiality, robustness across neural-network architectures, and threat models reflecting real-world settings.The section also calls for studying the correctness and confidence of XAI methods against black-box attacks.Source: F. XAI Security and Privacy · p. 23

G. Accountable and Systematic Decision Making

  • Robust XAI paradigms and their deployment and testing in crucial use cases are anticipated for systematic automated decision-making.The section connects XAI tools with system requirements, measurements, and periodic user testing.Source: G. Accountable and Systematic Decision… · p. 23

VI. CONCLUSION

  • The survey reviews XAI solutions, IoT use cases, adaptive architectures, and research challenges, aiming to outline gaps and inform future work.Its coverage includes 5G, cloud services, big-data management, IIoT, and other smart applications.Source: §VI · p. 23

Related papers

Loading 2211.01036v2…