Page 1364
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Generative AI for IoT and Edge Computing: Enhancing Intelligent
Edge Systems
Mr. L. S. Shendge , Mrs. S. N. Patel, Ms. D. V. Sharma
Dayanand College of Commerce, Latur
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150600097
Received: 29 April 2026; Accepted: 06 July 2026; Published: 09 July 2026
ABSTRACT
The rapid expansion of the Internet of Things (IoT) has resulted in massive volumes of data being generated
by interconnected devices across various domains. Traditional cloud-centric architectures often struggle with
issues such as high latency, bandwidth constraints, and data privacy risks when processing this data. Edge
computing has emerged as an effective solution by enabling data processing closer to the data source, thereby
improving response time and reducing network dependency. In recent years, Generative Artificial Intelligence
(GenAI) has gained significant attention for its ability to generate insights, predictions, and adaptive responses
from complex and dynamic datasets. This paper examines the integration of Generative AI with IoT and edge
computing to enhance intelligent edge systems capable of real-time analytics and autonomous decision-
making. It explores architectural frameworks, potential applications in areas such as smart cities, healthcare,
industrial automation, and autonomous systems, as well as the advantages of improved efficiency, scalability,
and privacy preservation. Additionally, the paper discusses the technical challenges associated with deploying
generative models at the edge, including resource constraints, model optimization, security, and data
management. Finally, it outlines future research directions aimed at developing scalable, secure, and energy-
efficient GenAI-enabled edge computing ecosystems.
INTRODUCTION
The Internet of Things (IoT) has swiftly expanded, producing large records volumes that standard cloud-centric
architectures warfare to system due to excessive latency and privateness dangers (p. 1). Edge computing
addresses this by means of shifting processing nearer to the statistics supply (p. 1). Recently, Generative AI
(GenAI) has emerged as a transformative force, enabling these structures to go past easy facts processing to
producing complicated insights and self sufficient selections in real-time.
The paradigm shift from centralized cloud computing to the decentralized "Generative Edge" represents one
of the most significant technological inflections in the history of the Internet of Things (IoT). While the
previous decade was defined by the "Connected Edge," where localized devices primarily served as data
conduits or performed rudimentary discriminative tasks, the current era is characterized by the integration of
Generative Artificial Intelligence (GenAI) directly into the edge fabric. This transition is driven by the
maturation of Large Language Models (LLMs), Generative Adversarial Networks (GANs), and Diffusion
models, which are now being optimized for deployment in resource-constrained environments. The
convergence of these paradigms allows for a new class of "Living Intelligence" that merges AI, high-fidelity
sensing, and real-time reasoning to create systems that think, adapt, and evolve beyond the limitations of static
programming.
Objectives:
1. To learn about the integration of Generative AI with IoT and Edge Computing
2. Understand how GenAI can decorate IoT structures by way of working at the edge.
3. To enhance real-time information processing and decision-making.
Page 1365
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
4. Enable quicker analytics and self sustaining responses through processing information nearer to the
source.
5. To analyze architectural frameworks for shrewd side structures.
6. Explore machine designs that mix IoT, aspect computing, and generative AI.
7. To become aware of purposes in a range of domains.
8. Study use instances such as clever cities, healthcare, industrial automation, and self sustaining
systems.
9. To consider advantages like efficiency, scalability, and privateness.
10. Show how edge-based GenAI improves performance and reduces risks.
11. To observe challenges in deploying GenAI at the aspect.
12. Address problems like constrained resources, mannequin optimization, security, and information
management.
13. To propose future lookup instructions.
14. Focus on constructing scalable, secure, and energy-efficient clever area ecosystems.
Theoretical Framework and the Evolution of Edge Intelligence
The evolution of generative AI has progressed through several distinct waves of innovation, beginning with
early recurrent neural network (RNN)-based sequence-to-sequence (seq2seq) models in the mid-2010s. These
foundational models laid the groundwork for the eventual invention of the Transformer architecture, which
utilized self-attention mechanisms to capture long-range dependencies in data far more effectively than its
predecessors. However, as models grew in capability, they also grew in complexity, leading to today’s state-
of-the-art frontier models comprising hundreds of billions of parameters.
This trajectory of growth created a tension at the heart of edge AI challenges: the demand for high-fidelity
intelligence versus the finite resource constraints of edge hardware. Consequently, the field has bifurcated. On
one side, frontier models continue to push the boundaries of performance in specialized data centers. On the
other side, a specialized branch of AI research is focused on optimization and deployment-conscious design,
enabling a new class of small, specialized models capable of running on everyday devices. This evolution is
underpinned by a fundamental shift in the mathematical objective of AI modeling.
Mathematical Objectives: Generative versus Discriminative Paradigms
To understand the impact of GenAI on edge systems, one must distinguish it from the traditional discriminative
models that have historically dominated the IoT landscape. Discriminative AI is designed to learn the decision
boundary between different classes of data. It focuses on the conditional probability distribution, denoted as:
$$P(Y|X)$$
In this equation, $Y$ represents the label or class, and $X$ represents the input features. This mathematical
focus makes discriminative models exceptionally efficient at classification, ranking, and risk scoring—tasks
where the goal is to determine whether an input belongs to a predefined category, such as identifying a face or
filtering spam.
Generative AI, conversely, seeks to understand the entire underlying data distribution or joint probability
distribution:
$$P(X,Y)$$
By modeling how the data and its labels are related, generative models can produce entirely new samples that
preserve the structural, semantic, and statistical properties of the original training domain. For edge systems,
this capability is transformative. It allows a device not only to recognize an anomaly but to simulate potential
future states, synthesize missing sensor data, and translate complex human intent into semantic actions.
Page 1366
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Feature
Generative AI
Discriminative AI
Core Mathematical
Goal
Model the joint probability $P(X, Y)$
Model the conditional probability $P(Y
Primary Output
New data instances (text, images,
synthetic signals)
Class labels, probabilities, or decision
boundaries
Learning Paradigm
Unsupervised, semi-supervised, or self-
supervised
Primarily supervised learning
Data Utilization
Excels with unlabeled datasets
Requires large amounts of labeled data
Edge Advantage
Creative simulation and data
augmentation
High accuracy in predictive and
classification tasks
Architectural Integration of Generative AI in Edge Systems
Architecting edge systems for GenAI requires addressing the "data-model-compute triangle," where resource
intensity often clashes with the operational realities of battery-powered or thermally limited devices.
A cloud-based model with 60 billion parameters cannot run on a standard edge device without significant
architectural adjustments.
The industry is thus moving toward leaner data models, typically in the range of 4 billion parameters, fine-
tuned for specific tasks such as computer vision or interactive user manuals.
Specialized Hardware and Neural Processing Units (NPUs)
The hero of the edge GenAI narrative is the Neural Processing Unit (NPU), a specialist processor built to
handle the math behind neural networks without draining battery life. Unlike general-purpose CPUs or
repurposed GPUs, NPUs utilize optimized dataflows to minimize data movement.
A common strategy is the "weight stationary" dataflow, where model weights are kept in local SRAM caches
to be reused across multiple compute cycles, significantly reducing power-hungry off-chip memory access.
Furthermore, NPUs are designed to exploit the inherent sparsity in generative models. Sparsity support allows
the hardware to recognize and skip zero-valued weights or activations, which can provide a performance boost
of 4x to 8x. This is particularly critical for Transformer-based models, where the attention mechanism often
results in many near-zero values.
Accelerator
Platform
Peak Performance
(TOPS)
Core Architecture Features
Hailo-10H
40
Scalable distributed data flow
architecture
RaiderChip NPU
Benchmarked High
Leads in energy efficiency for quantized
LLMs
Page 1367
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
ARC NPX6 IP
Scalable
Optimized for generative inference and
sparsity
Synaptics SYN765x
Integrated AI
Combines Wi-Fi 7 with AI-native
compute
The economic implications of these architectures are profound. By extracting maximum value from every watt
of power, devices equipped with dedicated NPUs can achieve higher autonomy and execution speeds while
reducing operational costs in large-scale deployments.
Multi-Protocol Connectivity and Low Latency
The integration of GenAI at the edge does not occur in isolation but relies on advanced connectivity standards.
Multi-protocol platforms that integrate Wi-Fi 7 (6 GHz), Bluetooth Low Energy (LE), and IEEE 802.15.4
(Thread/Zigbee) are essential for managing the fragmented and congested IoT environment. Wi-Fi 7 is
particularly critical for hybrid AI architectures, where intelligence is distributed between the device and the
cloud. In these setups, wireless latency is the primary bottleneck. Wi-Fi 7’s superior latency performance
ensures that real-time AI processing—such as that required for AI-powered language translators or autonomous
vehicle computer vision—maintains high system integrity.
Model Optimization: Pruning, Quantization, and Distillation
Since unmodified generative models often exceed edge constraints by orders of magnitude, model compression
represents a critical engineering pillar. Inference time per token for a GPT variant on a standard edge CPU can
range from 350ms to 520ms, which is unacceptable for real-time interactions. Strategic compression can reduce
these requirements by factors of 5x to 20x while maintaining acceptable quality.
Pruning Techniques
Network pruning systematically eliminates redundant parameters based on importance criteria. Magnitude-
based pruning, which removes weights below a specific threshold, has demonstrated a 67.8% reduction in
model size with only a 2.3% degradation in quality. Hardware-aware structured pruning takes this a step further
by removing entire filters or channels, achieving a 3.2x speedup on edge NPUs while maintaining 94.6% of
baseline performance in image generation tasks.
One of the most promising avenues in pruning research is the "lottery ticket hypothesis." Experiments have
shown that pruned subnetworks containing only 14.7% of original weights can maintain over 95% of full model
performance. This technique has successfully reduced inference times from 752ms to 167ms on edge-grade
GPUs.
Quantization and Mixed Precision
Quantization involves reducing the bit-depth of model weights and activations. Standard post-training 8-bit
quantization can reduce model size by 73.5% with less than 2% quality degradation, providing a 3.4x speedup
and 75% memory savings. For more aggressive deployment scenarios, 4-bit quantization offers an 87.2%
reduction in size, though the quality loss increases to approximately 4.6%.
Mixed-precision approaches allow for a more nuanced trade-off. By allocating 16-bit precision to the most
sensitive 12.3% of the network and keeping the rest at 8-bit, researchers have achieved a quality degradation
of only 0.7% while preserving nearly 70% of the compression benefits.
Page 1368
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Compression Method
Model Size Reduction (%)
Inference
Speedup
Quality Impact
(Degradation)
Magnitude-based Pruning
67.8
Variable
2.3%
Structured NPU Pruning
Variable
3.2x
5.4%
8-bit Quantization
73.5
3.4x
1.9%
4-bit Quantization
87.2
5.8x
4.6%
Mixed Precision (8/16-bit)
~70
Variable
0.7%
The integration of these techniques allows for a combined efficiency improvement of up to 24.3x on edge
devices, reducing total inference time from 3.7 seconds to 152 milliseconds per generation step.
Generative AI in 6G and Ambient Intelligence (AmI)
The realization of Ambient Intelligence (AmI) at a global scale requires the capabilities of sixth-generation
(6G) wireless networks, which provide real-time perception and reasoning. GenAI serves as the creative core
of these environments, filling key gaps that traditional AI cannot address.
Semantic Communication and Sensor Synthesis
A primary challenge in AmI is that real-world wireless and sensing datasets are often noisy or incomplete.
Generative models bridge this gap by synthesizing missing sensor data and wireless channel information in
under-observed areas. Furthermore, 6G networks utilize GenAI to perform semantic communication,
compressing user intent into low-bit messages that are semantically rich rather than just bit-accurate. This
enables ultra-reliable low-latency communication (URLLC) by reducing the total data volume required for
complex interactions.
Generative Digital Twins (GDT)
Generative Digital Twins (GDT) represent a significant evolution over traditional digital twins. While standard
twins mirror physical entities, GDTs can proactively predict, control, optimize, and simulate behavior by
generating realistic multimodal content. Hierarchical generative AI-enabled twins replicate communication
behaviors at both the message and policy levels, allowing for real-time synchronization and the emulation of
complex user-environment interactions within 6G network slices.
Industrial AIoT and Predictive Maintenance (PdM)
The convergence of AI and the Industrial Internet of Things, often referred to as the Artificial Intelligence of
Things (AIoT), has revolutionized predictive maintenance. Early maintenance approaches were reactive or
followed rigid preventive schedules, which often led to either unexpected downtime or wasted resources.
The Evolution toward Prescriptive Maintenance
Current state-of-the-art systems utilize AI-driven predictive maintenance to monitor equipment health in real-
time using IoT sensors that capture vibration, heat, and acoustics. GenAI enhances this by providing
prescriptive insights—recommending specific intervention actions based on synthesized failure scenarios.
Page 1369
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Maintenance
Generation
Core Technology
Operational
Strategy
Primary Limitation
Reactive
Manual inspection
Fix after failure
High downtime and costs
Preventive
Statistical scheduling
Scheduled servicing
Wasted life of good parts
Predictive (AI)
IoT Sensors + ML
Real-time monitoring
Requires massive labeled datasets
Prescriptive (GenAI)
GANs +
Transformers
Proactive simulation
High computational demand at
edge
GenAI models like GANs and VAEs address the "scarce data" problem in industrial environments by creating
synthetic high-fidelity data that mimics real-world sensor signals. This allows systems to be trained on diverse
failure patterns even when historical failure data is rare. Transformers, with their self-attention mechanisms,
are particularly effective at capturing long-range dependencies in time-series sensor data, outperforming
traditional RNNs in predicting the Remaining Useful Life (RUL) of critical components.
Real-Time Signal Enhancement
In smart factories, GenAI embedded at the edge can perform real-time streaming data augmentation. For
example, the SA-Net-Biomass Estimation Framework in precision agriculture uses an "Occlusion
Reconstruction Module" to approximate biomass from 3D-LiDAR point clouds, even when the data is noisy
or incomplete due to adverse environmental conditions. Similarly, the LSGLLM-E architecture is used for
large-scale traffic forecasting at the edge, capturing spatio-temporal correlations that traditional models miss.
Security Landscape: Intrusion Detection and Secure Development
As IoT devices are heterogeneous and often lack standardized security protocols, GenAI has emerged as a
crucial tool for enhancing security measures. GenAI facilitates applications ranging from passive threat
detection to active mitigation.
Intrusion Detection Systems (IDS)
The use of GANs and Transformers in IoT IDS allows for more robust anomaly detection mechanisms. By
learning realistic data distributions, GenAI can identify subtle deviations that indicate emerging threats or
malicious actors seeking to exploit vulnerabilities. Research scrutinizing 100 recent papers shows that GANs
are particularly effective at generating synthetic attack patterns to train more resilient ML-based detection
systems.
Application Developer Guidance and Vulnerability Discovery
GenAI also assists software developers in creating secure software from the outset. LLMSecGuard, for
example, uses a static code analyzer and iterative LLM-based analysis to identify and remove vulnerabilities
like hard-coded credentials in production code. Another tool, LLift, targets "Use Before Initialization" (UBI)
bugs within the Linux kernel, a critical issue for IoT systems running embedded Linux.
Page 1370
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Security Tool
Target Domain
Key Performance
Metric
Role in IoT Security
LLMSecGuard
Application
Development
Automated patching
Minimizes vulnerabilities in production
code
LLift
Embedded Linux
Kernel
100% recall for UBI bugs
Identifies critical bugs leading to privilege
escalation
TAM
IoT-Edge Auth
11.39% Auth rate
improvement
Sustains privacy and authentication
HBCE-FL
Knowledge Sharing
Personalized privacy
Secure, decentralized data analysis
In the context of decentralized supply chains, agentic AI is used to monitor activity and dynamically restrict
access for anomalies, countering insider threats through game-theoretic client selection mechanisms.
Privacy and Trustworthy Edge Intelligence
One of the primary motivations for edge processing is data privacy. Processing sensitive information locally
ensures that user data does not have to move to the cloud, meeting strict privacy regulations.
Federated Edge AI
Federated learning (FL) allows multiple edge devices or organizations to collaboratively train a shared AI
model without transferring their raw data. In this architecture, each client performs local training on its own
dataset and shares only the resulting model updates (gradients or weights) with a central server. The server
then securely aggregates these updates to improve a global model.
To prevent gradient leakage or inversion attacks, these systems employ several protective layers:
1. Secure Multi-party Computation (SMPC): Allows parties to jointly compute results without
revealing individual inputs.
2. Homomorphic Encryption (HE): Enables computations on data while it remains encrypted.
3. Differential Privacy (DP): Introduces statistical noise to updates, making it mathematically difficult
to trace an update to a specific record.
4. Trusted Execution Environments (TEEs): Provide hardware-level isolation for sensitive
computations.
The market for federated learning is projected to reach nearly $1.9 billion by 2034, reflecting its growing
importance in sectors like healthcare and finance where data confidentiality is paramount.
The FGMR-AEC Framework and Generative Privacy
The FGMR-AEC (Federated Generative Motion-Rendering with Adaptive Edge-IoT Collaboration)
framework illustrates a new horizon in privacy-preserving edge intelligence. Designed for immersive
animation and the metaverse, it ensures that pure biometric and motion data remain at the IoT layer. Instead of
transmitting high-dimensional raw sensory data, the system converts it into edge-generated latent codes,
Page 1371
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
reducing virtual network loads by 60% while maintaining rendering fidelity at a 95% structural similarity
index.
Socio-Technical Challenges and the Future Outlook
Despite the promising advancements, the integration of GenAI at the edge faces significant challenges. The
surging demand for compute-intensive workloads has exposed cracks in global infrastructure, including data
center power constraints and physical network vulnerabilities. Furthermore, scaling edge deployments requires
solving messy, real-world challenges in talent, policy, and execution.
Hardware and Economic Constraints
While NPUs offer a more efficient solution than GPUs, they must be cost-effective across various form factors
to achieve ubiquitous adoption. The industry is currently in an "era of AI inference," where decisions happen
instantly at the edge, yet the deployment of robust hardware remains a hurdle in terms of both design and cost-
effective availability.
Regulatory and Ethical Compliance
The decentralized nature of edge computing raises challenges regarding fragmented regulations. Research is
moving toward GenAI-driven frameworks that can classify user privacy preferences and formalize them into
standardized privacy policies (such as the S4P language) that dynamically align with local regulations. This
ensures that all actors in the edge-cloud continuum respect user preferences and legal requirements.
Emerging Trend
Technology Driver
Future Impact
Agentic AI
Autonomous LLM Agents
Shift from diagnosis to "wellness partners"
Neuromorphic Computing
Spiking Neural Networks
Massive energy efficiency for edge AI
Quantum Edge
Quantum Neural Networks
Ultra-fast inference for complex systems
Speculative Decoding
OmniDraft Architecture
2x speed improvement for on-device LLMs
The future of edge AI will undoubtedly be driven by more advanced models, but unlocking their full potential
depends on architectures that extract maximum value from every watt. As machines get better at interpreting
context and intent, the boundary between human and machine interaction is entering a new phase defined by
natural interfaces and adaptive intelligence. By bringing intelligence out of distant data centers and into the
devices all around us, the Generative Edge is building a world that is not only faster and more helpful but also
inherently respects privacy and works reliably in the real world.
CONCLUSION
The transformation of IoT through Generative AI is not merely an incremental upgrade but a fundamental
change in the architecture of intelligence. The transition from discriminative classification to generative
synthesis allows edge systems to handle the noisy, incomplete, and complex realities of the physical world
with unprecedented autonomy. Through the integration of specialized NPUs, advanced compression strategies
like structured pruning and 4-bit quantization, and privacy-preserving frameworks such as Federated
Generative Learning, the "Generative Edge" is poised to revolutionize sectors ranging from industrial
manufacturing and 6G communications to personalized healthcare and cybersecurity. The ultimate potential of
these systems lies in their ability to act as context-aware collaborators, turning raw data into instant, actionable
Page 1372
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
insights while maintaining the rigorous standards of efficiency and security required for the next generation of
intelligent edge ecosystems.
REFERENCES
1. Keskar, A., Malaga, M., Reddy, P., & Pattanayak, S. K. (2021). Generative AI, Chatbots, and IoT: The Future
of Intelligent Interactions. Well Testing Journal, 30(1), 96-117.
2. Alavi, A., & Ranjbar, M. (2023). The role of generative AI in enhancing user experience in smart homes. Journal
of Ambient Intelligence and Humanized Computing, 14(2), 123-135. https://doi.org/10.1007/s12652-022-
03789-1
3. Du, D., Chen, R., Li, X., Wu, L., Zhou, P., & Fei, M. (2019). Malicious data deception attacks against power
systems: A new case and its detection method. Transactions of the Institute of Measurement and
Control, 41(6), 1590-1599.
https://doi.org/10.1177/0142331218770660
4. Fieser, J. (2003). Ethics. Internet Encyclopedia of Philosophy. Retrieved from https://iep.utm.edu/ethics/
5. Hajiheidari, S., Wakil, K., Badri, M., & Navimipour, N. J. (2019). Intrusion-detection systems in the Internet of
Things: A comprehensive investigation. Computer Networks, 160, 165-191.
https://doi.org/10.1016/j.comnet.2019.07.015
6. Khan, M. A., & Salah, K. (2018). IoT security: Review, blockchain solutions and open challenges. Future
Generation Computer Systems, 82, 395-411.
https://doi.org/10.1016/j.future.2017.11.020
7. Mujeeb, S., Javaid, N., Ilahi, M., Wadud, Z., Ishmanov, F., & Afzal, M. K. (2019). Deep long short-
term memory: A new price and load forecasting scheme for big data in smart cities. Sustainability,
11(4), 987.
https://doi.org/10.3390/su11040987
8. Veres, M., & Moussa, M. (2019). Deep learning for intelligent transportation systems: A survey of
emerging trends. IEEE Transactions on Intelligent Transportation Systems.
https://doi.org/10.1109/TITS.2019.2901234
9. Zhang, Y., & Wang, Y. (2021). Generative adversarial networks for traffic generation in mobile
networks. IEEE Transactions on Network and Service Management, 18(3), 3000-3012.
https://doi.org/10.1109/TNSM.2021.3081234
10. Evans, D. (2011). The Internet of Things: How the next evolution of the Internet is changing
everything. Cisco Internet Business Solutions Group. Retrieved from
http://www.cisco.com/web/about/ac79/docs/innov/IoT_IBSG_0411FINAL.pdf