Chang Liu

I am a PhD student at Syracuse University, advised by Kristopher Micinski.

My research is on reverse engineering, at the intersection of programming languages, machine learning, and security. I build systems that recover interpretations a human or a language model can actually reason about from stripped binaries.

My work spans the evolution of reverse engineering, from machine-learning-based techniques to modern approaches built around LLMs.

• Assemblage, a distributed build system that compiles open-source software at scale into families of binary datasets, NeurIPS Datasets & Benchmarks 2024. Its successor, Assemblage-DeepHistory, adds temporal coverage, with CVE labels and multi-year build history.
• Manifold, a declarative decompiler that treats decompilation the way modern compilers treat compilation: a sequence of small, logic-defined passes over a shared, monotonically growing fact store, carrying ambiguous liftings forward as parallel candidates with provenance and resolving them in a final selection phase.
• My ongoing work explores behavioral evaluation of LLM-based decompilers, LLM-assisted decompilation, and symbolic decompilation.

I work in Rust, Python, Datalog, and C. My experience spans the CompCert and LLVM toolchains, as well as Ghidra, IDA Pro, and AFL++. My interests span decompiler construction, vulnerability analysis, and logic programming.