Joonghyuk Hahn

About Me

I am a researcher dedicated to building provably safe, verifiable, and trustworthy AI systems. I earned my Ph.D. in Computer Science from Yonsei University under the supervision of Prof. Yo-Sub Han. My dissertation, titled "Symbolic Techniques for Deep Learning in Natural Language Processing and Program Analysis," explores the intersection of formal methods and modern neural models.

My research vision is to bridge formal language theory and AI/NLP. By applying theoretical tools like formal grammars and automata, I create frameworks where AI behavior can be formally specified and mathematically verified, moving beyond heuristics to rigorous guarantees.

Research Vision & Interests

• Verifiable AI Safety: Formalizing safety constraints as decidable properties to develop systems with mathematical verification of outputs.
• Symbolic-Neural Integration: Bridging symbolic methods (logical constraints, decidability) with neural models for compositional generalization and systematic problem-solving.
• Constrained & Controllable Generation: Developing grammar-based frameworks for structured, safe generation in code synthesis and mathematical reasoning.
• Mechanistic Interpretability: Using automata theory to understand the internal circuits and reasoning processes of LLMs.
• Neuro-Symbolic Program Analysis: Structure-aware semi-supervised learning for reliable code summarization and complexity prediction.

Recent News