VulnArmor: mitigating software vulnerabilities with code resolution and detection techniques

VulnArmor: mitigating software vulnerabilities with code resolution and detection techniques

연구 분야: Strategies

학회: International Journal of Information Technology

In today’s swiftly evolving digital environment, the security and dependability of software applications are crucial. In light of industries’ increasing reliance on software, identifying and mitigating vulnerabilities is essential for protecting data, systems, and user trust. With data-driven methodologies, there is increased interest in using Artificial Intelligence (AI) and Machine Learning (ML) for software assurance to construct trustworthy software systems. This research addresses the urgent need for an automated and comprehensive approach to code resolution and vulnerability detection, providing a robust solution to improve software security and reduce potential risks. Code resolution is implemented by fine-tuning Large Language Models (LLM) like Generative Pre-Trained Transformers (GPT)-2, Text-to-Text Transfer Transformers (T5), Bidirectional Encoder Representations from Transformers (BERT), and Large Language Model Meta AI (LLaMA). Secondly, vulnerable code detection plays a crucial role in evaluating the correctness of resolved code and identifying any remaining vulnerabilities. This essential step not only validates the efficacy of code resolution but also identifies areas where additional mitigation efforts are required. Utilizing Deep Learning (DL) models, the top performer of the study, Convolutional Neural Network (CNN), achieved a remarkable 93% accuracy rate, demonstrating its prowess in protecting software applications against potential attacks.