Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence

Cognitive Computation, Journal Year: 2023, Volume and Issue: 16(1), P. 45 - 74

Published: Aug. 24, 2023

Abstract Recent years have seen a tremendous growth in Artificial Intelligence (AI)-based methodological development broad range of domains. In this rapidly evolving field, large number methods are being reported using machine learning (ML) and Deep Learning (DL) models. Majority these models inherently complex lacks explanations the decision making process causing to be termed as 'Black-Box'. One major bottlenecks adopt such mission-critical application domains, banking, e-commerce, healthcare, public services safety, is difficulty interpreting them. Due rapid proleferation AI models, explaining their getting harder which require transparency easy predictability. Aiming collate current state-of-the-art black-box study provides comprehensive analysis explainable (XAI) To reduce false negative positive outcomes back-box finding flaws them still difficult inefficient. paper, XAI reviewed meticulously through careful selection research. It also in-depth evaluation frameworks efficacy serve starting point for applied theoretical researchers. Towards end, it highlights emerging critical issues pertaining research showcase major, model-specific trends better explanation, enhanced transparency, improved prediction accuracy.

Language: Английский

---

AI-Empowered Fog/Edge Resource Management for IoT Applications: A Comprehensive Review, Research Challenges, and Future Perspectives

IEEE Communications Surveys & Tutorials, Journal Year: 2023, Volume and Issue: 26(1), P. 619 - 669

Published: Nov. 30, 2023

The proliferation of ubiquitous Internet Things (IoT) sensors and smart devices in several domains embracing healthcare, Industry 4.0, transportation agriculture are giving rise to a prodigious amount data requiring everincreasing computations services from cloud the edge network.Fog/Edge computing is promising distributed paradigm that has drawn extensive attention both industry academia.The infrastructural efficiency these paradigms necessitates adaptive resource management mechanisms for offloading decisions efficient scheduling.Resource Management (RM) non-trivial issue whose complexity result heterogeneous resources, incoming transactional workload, node discovery, Quality Service (QoS) parameters at same time, which makes efficacy resources even more challenging.Hence, researchers have adopted Artificial Intelligence (AI)-based techniques resolve abovementioned issues.This paper offers comprehensive review issues challenges Fog/Edge by categorizing them into provisioning task offloading, scheduling, service placement, load balancing.In addition, existing AI non-AI based state-of-the-art solutions been discussed, along with their QoS metrics, datasets analysed, limitations challenges.The survey provides mathematical formulation corresponding each categorized issue.Our work sheds light on research directions cutting-edge technologies such as Serverless computing, 5G, Industrial IoT (IIoT), blockchain, digital twins, quantum Software-Defined Networking (SDN), can be integrated frameworks fog/edge-of-things improve business intelligence analytics amongst IoT-based applications.

Language: Английский

---

Internet of Things and Wireless Sensor Networks for Smart Agriculture Applications: A Survey

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 145813 - 145852

Published: Jan. 1, 2023

The increasing food scarcity necessitates sustainable agriculture achieved through automation to meet the growing demand. Integrating Internet of Things (IoT) and Wireless Sensor Networks (WSNs) is crucial in enhancing production across various agricultural domains, encompassing irrigation, soil moisture monitoring, fertilizer optimization control, early-stage pest crop disease management, energy conservation. application protocols such as ZigBee, WiFi, SigFox, LoRaWAN are commonly employed collect real-time data for monitoring purposes. Embracing advanced technology imperative ensure efficient annual production. Therefore, this study emphasizes a comprehensive, future-oriented approach, delving into IoT-WSNs, wireless network protocols, their applications since 2019. It thoroughly discusses overview IoT WSNs, architectures summarization protocols. Furthermore, addresses recent issues challenges related IoT-WSNs proposes mitigation strategies. provides clear recommendations future, emphasizing integration aiming contribute future development smart systems.

Language: Английский

---

Revolutionizing Cyber Threat Detection With Large Language Models: A Privacy-Preserving BERT-Based Lightweight Model for IoT/IIoT Devices

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 23733 - 23750

Published: Jan. 1, 2024

The field of Natural Language Processing (NLP) is currently undergoing a revolutionary transformation driven by the power pre-trained Large Models (LLMs) based on groundbreaking Transformer architectures. As frequency and diversity cybersecurity attacks continue to rise, importance incident detection has significantly increased. IoT devices are expanding rapidly, resulting in growing need for efficient techniques autonomously identify network-based networks with both high precision minimal computational requirements. This paper presents SecurityBERT, novel architecture that leverages Bidirectional Encoder Representations from Transformers (BERT) model cyber threat networks. During training we incorporated privacy-preserving encoding technique called Privacy-Preserving Fixed-Length Encoding (PPFLE). We effectively represented network traffic data structured format combining PPFLE Byte-level Byte-Pair (BBPE) Tokenizer. Our research demonstrates SecurityBERT outperforms traditional Machine Learning (ML) Deep (DL) methods, such as Convolutional Neural Networks (CNNs) or Recurrent (RNNs), detection. Employing Edge-IIoTset dataset, our experimental analysis shows achieved an impressive 98.2% overall accuracy identifying fourteen distinct attack types, surpassing previous records set hybrid solutions GAN-Transformer-based architectures CNN-LSTM models. With inference time less than 0.15 seconds average CPU compact size just 16.7MB, ideally suited real-life suitable choice deployment resource-constrained devices.

Language: Английский

---

Deep learning for cyber threat detection in IoT networks: A review

Internet of Things and Cyber-Physical Systems, Journal Year: 2023, Volume and Issue: 4, P. 110 - 128

Published: Oct. 10, 2023

The Internet of Things (IoT) has revolutionized modern tech with interconnected smart devices. While these innovations offer unprecedented opportunities, they also introduce complex security challenges. Cybersecurity is a pivotal concern for intrusion detection systems (IDS). Deep Learning shown promise in effectively detecting and preventing cyberattacks on IoT Although IDS vital safeguarding sensitive information by identifying mitigating suspicious activities, conventional solutions grapple challenges the context. This paper delves into cutting-edge methods security, anchored Learning. We review recent advancements IoT, highlighting underlying deep learning algorithms, associated datasets, types attacks, evaluation metrics. Further, we discuss faced deploying suggest potential areas future research. survey will guide researchers industry experts adopting techniques detection.

Language: Английский

---

Improving IoT Security With Explainable AI: Quantitative Evaluation of Explainability for IoT Botnet Detection

IEEE Internet of Things Journal, Journal Year: 2024, Volume and Issue: 11(10), P. 18237 - 18254

Published: Jan. 31, 2024

Detecting botnets is an essential task to ensure the security of IoT systems. Machine learning-based approaches have been widely used for this purpose, but lack interpretability and transparency models often limits their effectiveness. In research paper, our aim improve high-performance machine learning botnet detection by selecting higher-quality explanations using explainable artificial intelligence (XAI) techniques. We three datasets induce binary multiclass classification detection, with Sequential Backward Selection employed as feature selection technique. then use two post hoc XAI techniques such LIME SHAP, explain behaviour models. To evaluate quality generated methods, we faithfulness, monotonicity, complexity, sensitivity metrics. ML in work achieve very high rates a limited number features. Our findings demonstrate effectiveness methods improving Specifically, applying SHAP XGBoost model yield Consistency, low sensitivity. Furthermore, outperforms achieving better results these

Language: Английский

---

Recent endeavors in machine learning-powered intrusion detection systems for the Internet of Things

Journal of Network and Computer Applications, Journal Year: 2024, Volume and Issue: 229, P. 103925 - 103925

Published: June 20, 2024

Language: Английский

---

Multi-objective optimization algorithms for intrusion detection in IoT networks: A systematic review

Internet of Things and Cyber-Physical Systems, Journal Year: 2024, Volume and Issue: 4, P. 258 - 267

Published: Jan. 1, 2024

The significance of intrusion detection systems in networks has grown because the digital revolution and increased operations. method classifies network traffic as threat or normal based on data features. Intrusion system faces a trade-off between various parameters such accuracy, relevance, redundancy, false alarm rate, other objectives. paper presents systematic review Internet Things (IoT) using multi-objective optimization algorithms (MOA), to identify attempts at exploiting security vulnerabilities reducing chances attacks. MOAs provide set optimized solutions for process highly complex IoT networks. This identification multiple objectives detection, comparative analysis their approaches, datasets used evaluation. show encouraging potential enhance conflicting detection. Additionally, current challenges future research ideas are identified. In addition demonstrating new advancements techniques, this study gaps that can be addressed while designing

Language: Английский

---

Image Denoising Techniques for Cybersecurity and Forensic Applications

Advances in information security, privacy, and ethics book series, Journal Year: 2024, Volume and Issue: unknown, P. 117 - 142

Published: Dec. 6, 2024

With the proliferation of digital evidence in cybersecurity and forensic investigations, image denoising has become essential for accurate analysis, where high-quality visuals are critical identifying threats verifying information. This chapter explores advanced AI-driven techniques denoising, emphasizing role deep learning, convolutional neural networks (CNNs), generative models to enhance clarity. By leveraging artificial intelligence, these methods adaptively reduce noise while preserving features, improving both efficiency reliability processes. Our examination includes a comparative analysis traditional versus AI-based approaches, assessing their applicability effectiveness within environments. ultimately aims provide comprehensive overview cutting-edge AI that refine quality, supporting better decision-making complex, data-rich scenarios.

Language: Английский

---

Ethical Considerations in Drone Cybersecurity

Advances in information security, privacy, and ethics book series, Journal Year: 2024, Volume and Issue: unknown, P. 42 - 87

Published: Jan. 26, 2024

The rapid proliferation of drones, coupled with their increasing integration into various aspects our lives, has brought to the forefront a myriad ethical considerations in realm cybersecurity. This chapter delves deep intricate web challenges surrounding drone cybersecurity, aiming provide comprehensive understanding this critical issue. introduction sets stage by highlighting essential role ethics emphasizing need for responsible decision-making an age where drones are omnipresent. It lays out scope, objectives, and key concepts research, underscoring contributions it makes field. core explores principles underpinning cybersecurity elucidates how these can be applied domain technology. authors delve delicate balance between security privacy, discussing implications data collection, retention, surveillance context drones.

Language: Английский

---