Holistic Debloating in the Age of LLM Technology
Abstract
Approved for Public ReleaseCurrent methodologies for debloating have demonstrated significant efficacy in eliminating superfluous components while preserving the essential functionality of debloated programs. Despite this progress, several issues plague these techniques, such as stability, integration with DevOps, and usability. Moreover, there is a notable lack of metrics to assess the performance of various debloating tools. Our approach capitalizes on Large Language Models (LLMs) to address these challenges. We begin by proposing a metric to evaluate the performance of debloating tools. Following this, we introduce a framework, SEALED, that uses LLMs to select the most effective debloating tool, subsequently facilitating automated DevOps integration. Our team comprises Principal Investigators with diverse expertise in program analysis, machine learning, and large language models.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 15, 2023
- Source ID
- N000142412049
Entities
People
- Somesh Jha
Organizations
- Office of Naval Research
- United States Navy
- University of Wisconsin System