Abhishek Manhas | 2026 I.S. Symposium
狈补尘别:听Abhishek Manhas
罢颈迟濒别:听Bridging the Semantic Gap in Decompilation
惭补箩辞谤蝉:听Mathematics; Computer Science
础诲惫颈蝉辞谤蝉:听John Musgrave; Joseph Smith
Decompilation converts executable binaries into readable source-like code, but recovered programs often lose semantic clarity because compiler optimizations remove important information, including structure, type intent, and high-level abstractions. This study explores whether large language models can improve decompiled output through semantic-aware optimization, simplifying control flow and reconstructing higher-level representations while maintaining correctness. A pipeline is developed that combines static decompiler signals with constrained LLM rewriting and lightweight verification checks. Experiments show improvements in readability, abstraction quality, and structural similarity to reference implementations, also highlighting limitations such as hallucinated logic and context sensitivity, motivating a guarded workflow for reliable LLM-assisted reverse engineering.
Posted in Symposium 2026 on May 1, 2026.
