Idioms: A Simple and Effective Framework for Turbo-Charging Local Neural Decompilation with Well-Defined Types
Luke Dramko
Network and Distributed System Security (NDSS) Symposium 2026 · Day 1 · Distributed Systems
This talk presents **Idioms**, a framework for improving neural decompilation -- using locally-hosted LLMs to recover readable, type-rich source code from compiled executables. The key insight is twofold: **user-defined types (UDTs) like structs and unions are integral to code structure** and must be predicted alongside the code, and **interprocedural analysis is essential** because type information often can only be inferred from how data is used across multiple functions. By including the bodies of related decompiled functions in the LLM's context and training it to predict both code and type definitions simultaneously, Idioms outperforms existing neural decompilers like **LLM4Decompile** on both correctness and code quality metrics.
AI review
A practically important advance in neural decompilation that solves two real problems: recovering user-defined types (structs/unions) and leveraging cross-function context. The hash table ablation -- where the model predicts the wrong algorithm without interprocedural context -- perfectly illustrates why type recovery requires whole-program information. Running on local 7B models makes this immediately useful for classified RE work. The dependency-graph metric is a genuine contribution to evaluation methodology.