Neuron-Anchored Rule Extraction for Large Language Models via Contrastive Hierarchical Ablation

Jan 1, 2026·
Francesco Sovrano
Gabriele Dominici
Marc Langheinrich
· 0 min read
Type
Publication
CoRR
publications
Francesco Sovrano
Authors
Francesco Sovrano
Postdoc
Francesco Sovrano is a computer scientist and data-science researcher focused on explainability for responsible AI. He earned a PhD in Data Science and Computation in 2023 from the University of Bologna, in association with the Polytechnic University of Milan, developing a computational theory of explanations with applications in user interfaces, regulatory compliance, and reinforcement learning. As a postdoctoral researcher at the University of Zurich he applied explanation theories to software engineering, AI in education, and EU regulation, while also researching machine learning for code. He later became an Early-Career Fellow at ETH Zurich’s Collegium Helveticum, creating XAI tools to reveal rules and biases in LLM-generated explanations. His work aims to identify and mitigate cognitive and statistical biases in human–AI interaction, advancing transparent, ethical, and trustworthy AI.
Gabriele Dominici
Authors
Gabriele Dominici
PhD Student
Gabriele Dominici is a PhD student at the Faculty of Informatics, Università della Svizzera italiana (USI). He works on developing interpretable machine-learning models that can generate counterfactuals, with the aim of improving the interaction between machines and humans.
Marc Langheinrich
Authors
Marc Langheinrich
Professor

Marc Langheinrich is a Full Professor for Computer Science at the Università della Svizzera italiana (USI) in Lugano, Switzerland, where he co-leads the People-Centered Computing Lab together with Prof. Silvia Santini. Marc received a masters degree (Diplom) in Computer Science from the University of Bielefeld, Germany, in 1997, and his PhD (Dr. sc.) from the ETH Zürich, Switzerland, in 2005. Starting in the fall of 1995, he spent a year as a Fulbright Scholar at the University of Washington, where he also completed his thesis work in the fields of information retrieval and software agents. In the fall of 1997 he joined NEC Research in Japan, where he worked on projects involving personalization and electronic commerce. From October 1999 to August 2008 he was a researcher at the Institute for Pervasive Computing at ETH Zürich, and he joined USI in September 2008.

Research interests

Marc’s main research interest lies at the intersection of privacy and ubiquitous computing. In a world full of invisible sensory systems, comprehensive digital dossiers could be easily and inconspicuously collected. This poses several challenges to the design of ubiquitous systems: Which communication protocols ensure adequate levels of privacy protection? How can user interfaces be structured to give people control over their personal information? And how feasible is comprehensive protection in a world of countless “smart” devices?

Professional service

  • Editor-in-Chief (2018-2023), IEEE Pervasive Computing
  • Editor, Dagstuhl Open Access Series in Informatics (OASIcs)
  • Director, Master in Management and Informatics, USI Lugano
  • Director, Master in Financial Technology and Computing, USI Lugano
  • Member, Equal Opportunities Committee, USI Lugano