In a digitalized world (“software is eating the world”), software quality in general and software security in particular play a central role.

One of our research goals is to contribute to the development of high-quality and secure software. An important means for this are intelligent methods to provide AI-enhanced coding assistance in developing new software, and also algorithms of program analysis, which automate the detection of various issues, including security vulnerabilities.

To this end, we focus on enhancing the capabilities of code-generating large language models (LLMs). This is evident in papers exploring the limitations of LLMs in code generation (like analyzing where they struggle) and evaluating their learning processes. These studies aim to improve LLMs’ effectiveness in generating and understanding code.

Furthermore, we also emphasizes theoretical frameworks and methodologies for program analysis. This is illustrated in the exploration of collaborative program analysis, modular soundness theories, and various frameworks for purity and cross-language analyses. These works seek to develop robust analytical methods to improve the reliability and consistency of code analysis across different programming languages and contexts.

Courses:

Recent Artificial Intelligence Papers:

  • Prompting neural-guided equation discovery based on residuals.
    Brugger, Pfanschilling, Richter, Mezini, Kramer, DIS 2025. 10.1007/978-3-032-05461-6_7
  • Integrating symbolic execution into the fine-tuning of code-generating LLMs.
    Sakharova, Anand, Mezini, NAACL 2025. 10.18653/v1/2025.naacl-srw.27
  • AI-assisted programming: From intelligent code completion to foundation models: A twenty-year journey.
    Mezini, SE 2025. 10.18420/se2025-02
  • A critical study of what code-LLMs (do not) learn.
    Anand, Verma, Narasimhan, Mezini, ACL 2024. 10.18653/v1/2024.findings-acl.939
  • Towards trustworthy AI software development assistance.
    Maninger, Narasimhan, Mezini, ICSE 2024. 10.1145/3639476.3639770

Recent Program Analysis Papers:

  • Unimocg: Modular call-graph algorithms for consistent handling of language features.
    Helm, Roth, Keidel, Reif, Mezini, SE 2025. 10.18420/se2025-15
  • Total recall? How good are static call graphs really?.
    Helm, Keidel, Kampkötter, Düsing, Roth, Hermann, Mezini, SE 2025. 10.18420/se2025-28
  • AXA: Cross-language analysis through integration of single-language analyses.
    Roth, Näumann, Helm, Keidel, Mezini, SE 2025. 10.18420/se2025-29
  • A modular soundness theory for the blackboard analysis architecture.
    Keidel, Helm, Roth, Mezini, ESOP 2024. 10.1007/978-3-031-57267-8_14
  • Unimocg: Modular call-graph algorithms for consistent handling of language features.
    Helm, Roth, Keidel, Reif, Mezini, ISSTA 2024. 10.1145/3650212.3652109