Logo image
Integrated Machine Learning Framework for Mitigation of Buffer Overflow, Software Supply Chain, and Adversarial Attacks
Dissertation   Open access

Integrated Machine Learning Framework for Mitigation of Buffer Overflow, Software Supply Chain, and Adversarial Attacks

Mst Shapna Akter
University of West Florida Libraries
Doctor of Philosophy (PHD), University of West Florida
2024

Metrics

766 File views/ downloads
56 Record Views

Abstract

Software security faces critical challenges in the digital age with risks from vulnerabilities likebuffer overflow, software supply chain threats, and adversarial attacks leading to severe outcomes such as data breaches and system compromises. Traditional detection and mitigation methods often fall short due to the evolving complexity of software systems, with conventional machine learning techniques struggling to keep pace with sophisticated modern threats. This highlights the untapped potential of quantum machine learning (QML). We introduce an integrated framework combining advanced machine learning (ML) and QML to bolster software security. Our approach includes neural networks like Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Generative Pre-trained Transformer 2 (GPT-2), enhanced with GloVe and fastText embeddings to analyze buffer overflow vulnerabilities in C and C++ code. We also propose a stacking ensemble model merging multiple neural architectures to improve the robustness of buffer overflow detection. Additionally, we developed Quantum Natural Language Processing (QNLP) for automated vulnerability detection, demonstrating that quantum LSTM (QLSTM) models enhance detection accuracy and efficiency. Our dissertation extends to software supply chain vulnerabilities, assessing the impact of machine learning and QML on varied dataset sizes. Results show that Quantum Neural Networks (QNNs), despite longer run times, achieve superior precision and recall, particularly in large datasets. We further investigate the resilience of ML and QML models against adversarial attacks, introducing the Quantum Fast Gradient Sign Method (QFGSM) and showcasing its effectiveness compared to traditional FGSM attacks. This underscores the resilience of QNNs and emphasizes the need for quantum-specific defense strategies.
pdf
Integrated Machine Learning Framework for Mitigation of Buffer Overflow, Software Supply Chain, and Adversarial Attacks3.43 MBDownloadView
Preprint Dissertation pdf Open Access

Details

Logo image