Vignali, Andrea (2025) AI-Empowered Cybersecurity in Cyber-Physical Systems through Natural Language Processing and Anomaly Detection. [Tesi di dottorato]

Documento PDF _TESI_PhD_Vignali_AI_Empowered_Cybersecurity_in_Cyber_Physical_Systems_through_Natural_Language_Processing_and_Anomaly_Detection.pdf Visibile a [TBR] Amministratori dell'archivio Download (8MB) | Richiedi una copia

Abstract

Natural Language Processing (NLP) has transitioned from a specialized field within computational linguistics to an integral aspect of Artificial Intelligence (AI), facilitating progress in various sectors, including biomedicine, software engineering, and cybersecurity. Despite its rapid progress, NLP still faces challenges related to data scarcity, computational cost, interpretability, and privacy, especially in specialized or high-risk domains where large-scale annotation and adaptation are difficult. This thesis develops along two main research threads that ultimately converge in cybersecurity. The first thread focuses on NLP and its domain-specific adaptations, addressing the limitations of data scarcity and imbalance in biomedical text processing and exploring the use of NLP models for software engineering tasks such as code translation and test case prioritization. The second thread investigates anomaly detection in complex Cyber-Physical Systems (CPSs) through multimodal deep learning models that integrate sensor and network data to improve the identification of system faults and attacks. These two directions merge in the final part of the thesis, where NLP and anomaly detection converge in cybersecurity applications. Specifically, a novel method is proposed that combines large language models (LLMs) and AI planning to extract structured knowledge from unstructured security data and automatically identify exploit chains, enabling proactive cyber defense. Overall, the thesis demonstrates how the reasoning capabilities of NLP and deep learning can enhance the detection, interpretation, and prevention of anomalies and threats in modern intelligent systems.

Downloads

Actions (login required)

Modifica documento