The manipulation of digital images has become a popular trend. Due to the development of image processing tools and visualization techniques integrating deep learning and artificial intelligence (AI) algorithms (in particular generative adversarial networks–GAN), this can pose serious threats to privacy and security. In recent years, Deepfake algorithms have been designed to exchange faces or modify facial features, potentially leading to more severe problems in this context. In this manuscript, we provide a comprehensive review of the two most important facial image processing technologies: (i) deepfake face manipulation; and (ii) face manipulation detection techniques. Furthermore, we explore the state-of-the-art of popular GAN techniques. In particular, three types of Deepfake face detection technologies are reviewed: (i) hand-crafted features (ii) artifacts; and (iii) learning-based features, while highlighting related improvements and challenges. Furthermore, this article discusses potential challenges and promising research directions for future investigation. We believe that this review has been organized to provide a structured analysis of important research papers and to discuss each study’s main findings and conclusions. Our investigation reveals shortcomings in manipulation detection benchmarks due to real-world scenario variations and biased dataset comparisons. Current research priorities revolve around enhancing GAN training stability, resolution, and manipulable facial features. Moreover, GANs have shown superior results in identifying fake images; however, their reliance prompts a systematic approach to detecting fakes. This dependency raises questions about detecting fake images with or without manipulated GAN architecture, urging the need for novel computational techniques to identify manipulations without GAN assistance.

Spiking neural networks (SNNs) use individual temporal spikes for computation and communication, simulating the actions of biological neurons. SNN had long been disregarded since it was thought to be intricate and difficult to analyze. We investigate the improvement of SpikeProp, a supervised learning model tailored for SNNs, in this work. Three distinct models are being proposed and investigated, including the proposed model 1, the proposed model 2, and the proposed model 3, each providing unique improvements to the SpikeProp algorithm. To accelerate convergence and adaptive learning rates, particle swarm optimization (PSO) and momentum factors are integrated into the proposed model 1. In proposed model 2, a rate dependency is introduced based on angle-driven learning. By incorporating PSO and learning rates, model 3 combines the strengths of both models 1 and 2. We believe, SNNs can be trained and classified more efficiently and accurately using these models. Furthermore, we examine how large language models (LLMs) might inform the design and interpretability of neural architectures and learning methodologies while also enhancing SNN training. Through the use of LLMs, we seek to enhance model transparency and encourage more Responsible AI (RAI) principles. A thorough evaluation and comparison of proposed models with traditional methods confirms that these models consistently outperform traditional methods for various real datasets. Consequently, they have a high potential for practical applications in neural network training in real-world settings and LLM-informed development, contributing to the advancement of AI systems.

Spiking neural networks (SNNs) mimic the functions of biological neurons by leveraging individual temporal spikes for communication and computation. Since SNN was perceived to be complex and analytically challenging, it had long been overlooked. In this study, we explore the enhancement of SpikeProp, a supervised learning model customized for SNNs. Three distinct models are being investigated, including the Proposed Model 1, the Proposed Model 2, and the Proposed Model 3, each providing unique improvements to the SpikeProp algorithm. To accelerate convergence and adapt learning rates, momentum factors are integrated into Proposed Model 1. In Proposed Model 2, a rate dependency is introduced based on Angle Driven Learning. By incorporating particle swarm optimization (PSO), Model 3 combines the strengths of Model 1 and 2. SNNs can be trained and classified more efficiently and accurately using these models. Furthermore, we examine how large language models (LLMs) might inform the design and interpretability of neural architectures and learning methodologies while also enhancing SNN training. Through the use of LLMs, we seek to enhance model transparency and encourage more Responsible AI (RAI) principles. A thorough evaluation and comparison of Proposed Model 1, Proposed Model 2, and Proposed Model 3 with traditional methods confirms that these models consistently outperform them. Consequently, they have a high potential for practical applications in neural network training in real-world settings and LLM-informed development, contributing to the advancement of AI systems.

The rise of the Industrial Internet of Things (IIoT) has brought a new era of automated communication to the industrial sector. However, increased connectivity and the use of low-power devices make IIoT networks highly vulnerable to security threats. This paper presents a hybrid intrusion detection system (IDS) for edge-based IIoT devices that uses two deep learning (DL) classifiers, the Sparse Auto Encoder (SAE) and the Bidirectional Long-Short Term Memory (BLSTM), to identify specific attack categories and provide early warning for security analysts. The SAE extracts patterns of the IIoT network through its sparse coding technique, while the BLSTM uses the output of the SAE to identify potential threats and intrusions. The proposed framework can pinpoint specific attack categories and aid security analysts in providing early warning and adopting proactive defensive methods. The edge-based approach improves routing flexibility and promotes interoperability between heterogeneous IIoT devices. The proposed framework achieved 99.61 % and 99.22 % accuracy under CICDDoS2019 and CICID2018 datasets and outperforms traditional and contemporary intrusion detection methods.