Journal of Cyber Security in Computer System

Federated Learning for Privacy-Preserving Artificial Intelligence Applications

Jagu Varalakshmi, Madem Meghana, P. Devi Sravanthi, T. Jagadeesh — Thu, 03 Sep 2026 00:00:00 +0000

Artificial Intelligence (AI) technology plays a significant role in the development of various solutions to address complex challenges in various sectors such as healthcare, banking, logistics, transport, and mobile computing, among others. Current artificial intelligence technologies rely on extensive amounts of data for the effective training of machine learning algorithms. Usually, data for training machine learning algorithms has been collected on a central server that performs the training of machine learning algorithms. However, collecting and storing data on the central server poses major threats to data privacy, security breaches, and legal obligations, especially for personal and organizational data that require protection. As such, in order to mitigate the above challenges, a new paradigm in machine learning algorithms referred to as federated learning has been developed. Under federated learning, machine learning models are usually trained at the client-side, which could include a smartphone, edge devices, or an organizational server. In this case, the client device would not transmit actual data to the central server but the parameters of the machine learning algorithms instead. Here, the central server would use different aggregation algorithms in the combination of the parameters of the machine learning models from multiple client devices. Communication cost would also be lowered because model parameters would only be communicated to the central server instead of actual data. More importantly, federated learning enables organizations to collaborate in building intelligent machines without violating any privacy and data protection rules.

Automated Prompt Optimization in Software Engineering: A Survey and Proposed Intent-Chaining Architecture

Praphthi H. P, Ramya B. P, Ranjan Prajwal, Sanjana B, Janaki K. B — Mon, 07 Sep 2026 00:00:00 +0000

Prompt engineering has emerged as a critical discipline for effectively utilizing Large Language Models (LLMs) across software engineering tasks, including code generation, debugging, documentation, and domain-specific language development. However, the rapid proliferation of prompting techniques has created a fragmented landscape, making it challenging for practitioners to identify suitable approaches for their specific requirements. This survey provides a comprehensive review and comparative analysis of contemporary prompt engineering methodologies in software engineering, examining eight representative studies across three key dimensions: prompt optimization techniques, evaluation frameworks, and domain-specific applications. They systematically analyze each approach based on its methodology, key contributions, and inherent limitations. Our analysis reveals several persistent challenges, including domain specificity, lack of interpretable evaluation feedback, reliance on single datasets, and fragmentation of complementary techniques. Based on these findings, critical research gaps were identified, and an integrated framework was proposed that combines intent extraction, few-shot prompting, and prompt chaining in a unified modular pipeline designed to overcome the limitations identified in existing approaches. This paper serves as both a comprehensive reference for researchers and practitioners in prompt engineering and a foundation for developing more accessible, interpretable, and generalizable LLM interaction tools for software engineering applications.

Aligning Academia with Industry 4.0: A Data-Driven Framework for Quantifying Graduate Readiness and Competency Gaps

Dattatraya Kumbhar, Akshaya Uttekar, Ashwini Kumbhar — Wed, 09 Sep 2026 00:00:00 +0000

The rapid convergence of cyber-physical systems, the Internet of Things (IoT), artificial intelligence, and big data analytics under Industry 4.0 has created a profound misalignment between higher education curricula and real-world industrial expectations. Traditional academic evaluation metrics rely on periodic qualitative reviews that fail to capture fast-evolving skill deficits across technical, cognitive, and socio-digital domains. To address this challenge, this paper proposes a quantitative, data-driven framework designed to systematically evaluate graduate readiness and map skill deficiencies against dynamic industrial requirements. By integrating multi-source data streams, including real-time labor market analytics, industry survey benchmarks, and Student Learning Outcome (SLO) metrics, the framework establishes a normalized Competency Readiness Index (CRI). This model applies a capability maturity mapping technique to evaluate performance across three primary domains: digital/technical foundations, collaborative problem-solving, and continuous learning adaptability. Empirical evaluation indicates that deploying this dynamic feedback model enables Higher Education Institutions (HEIs) to pinpoint curriculum vulnerabilities in real time, shorten curriculum revision cycles, and reduce post-graduation industry onboarding friction. Ultimately, this framework offers actionable insights for academic leaders, accreditation bodies, and industry partners seeking to align human capital development with the demands of the fourth industrial revolution.

Preserving Privacy in Federated Learning Systems through a Secure Multi-Layer Defense Framework against Security Threats

Mon, 05 Oct 2026 00:00:00 +0000

Although Federated Learning (FL) permits collaboration in model training with local data being kept private, model updates may be attacked by gradient leakage, model poisoning, membership inference, and eavesdropping. To combat these multi-vector attacks, the study proposes the Secure Multi-Layer Defense Framework (SMLDF), an architecture in which security measures are distributed across the three layers of client, communication, and server. Client-side security includes gradient clipping and Differential Privacy (DP), which limit the impact of individual samples and introduce noise to prevent reconstruction and inference attacks before updates are sent out of local machines. In the communication and summation layer, secure protocols and Secure Aggregation provide protection for gradient updates on the way. Anomaly detection and robust aggregation at the server side detect poisoned gradient updates from malicious clients. As per the qualitative analysis conducted, the framework will ensure a balance between the effectiveness of the model, resilience to multi-vector attacks, privacy, and communication. The framework is the foundation for future experiments using benchmarks such as FEMNIST and CIFAR-10. It compares model performance, multi-vector attack prevention capability, privacy guarantee, and communication cost with current FL security models. SMLDF sets up the foundation for future experiments on federated systems, where benchmark datasets such as FEMNIST and CIFAR-10 will be used in adversarial settings.

Cybersecurity Governance Model Toward Deepfake Manipulation

Comfort Ogechi Moses, Oluyomi Kolawole Akinyokun — Fri, 09 Oct 2026 00:00:00 +0000

This study developed and evaluated a Cybersecurity Governance Model (CGM) that integrates MesoNet-CNN-based deepfake detection with cybersecurity, data protection, and AI governance requirements. A sequential mixed-methods design combined quantitative deepfake detection with qualitative analysis of governance policies. The MesoNet-4 CNN, trained on the DFDC dataset using OpenCV, MTCNN, and class-weighted Binary Cross-Entropy loss under the Adam optimiser, achieved 94.43% accuracy, 95.2% precision, 93.7% recall, 94.4% F1-score, 0.981 AUC-ROC, and 0.883 MCC on 19,600 frames, outperforming XceptionNet, ResNet-50, and a baseline CNN. Governance requirements were mapped from ISO/IEC 27001:2022, NIST AI RMF, NIST CSF 2.0, GDPR, NDPA 2023, and the EU AI Act. A governance response function, G(x), classified detection probabilities into three tiers using thresholds of 0.41 and 0.72, cutting the governance-relevant missed detection rate by 13% and achieving 87.7% compliance coverage. The findings show that effective deepfake governance requires combining technical detection with structured, auditable policy responses, particularly within the Nigerian context.