Advancements in Large Language Models for Software Vulnerability Detection

The field of software vulnerability detection is rapidly advancing with the application of large language models (LLMs). Recent developments have focused on improving the performance and efficiency of LLMs in detecting vulnerabilities across multiple programming languages. This includes enhancing code preprocessing techniques to standardize code representation, exploring the feasibility of multilingual grammatical error correction, and leveraging data science insights for hardware security research. Notably, innovations in LLM-based software vulnerability assessment have shown promising results, with approaches such as in-context learning and information fusion demonstrating improved accuracy and effectiveness. Furthermore, research has highlighted the potential of sparse autoencoders as a lightweight and interpretable alternative for bug detection in Java functions. Noteworthy papers in this area include: Enhancing Large Language Models with Faster Code Preprocessing for Vulnerability Detection, which introduced SCoPE2, an enhanced version of the existing SCoPE framework with improved performance. A Preliminary Study of Large Language Models for Multilingual Vulnerability Detection, which evaluated the effectiveness of pre-trained language models and state-of-the-art LLMs across seven popular programming languages. Are Sparse Autoencoders Useful for Java Function Bug Detection, which explored the use of sparse autoencoders as a lightweight and interpretable alternative for bug detection in Java functions.

Sources

Enhancing Large Language Models with Faster Code Preprocessing for Vulnerability Detection

Exploring the Feasibility of Multilingual Grammatical Error Correction with a Single LLM up to 9B parameters: A Comparative Study of 17 Models

"vcd2df" -- Leveraging Data Science Insights for Hardware Security Research

A Preliminary Study of Large Language Models for Multilingual Vulnerability Detection

ICVul: A Well-labeled C/C++ Vulnerability Dataset with Comprehensive Metadata and VCCs

SVA-ICL: Improving LLM-based Software Vulnerability Assessment via In-Context Learning and Information Fusion

Designing and Contextualising Probes for African Languages

Are Sparse Autoencoders Useful for Java Function Bug Detection?

Built with on top of