AI models have developed rapidly and can now solve a variety of problems. But what exactly can they solve, and what can they not? In his doctoral dissertation, MSc Veeti Ahvonen approaches this question through mathematical logic, studying several important neural network architectures, including variants of the transformers behind large language models. The research provides exact matches between these architectures and logics: each logic captures precisely what problems its matching architecture can solve. The results also open promising directions for research into the explainability and verification of neural networks.
The doctoral dissertation titled On the Logical Foundations of Modern Computing Models by MSc Veeti Ahvonen will be publicly examined in the Faculty of Information Technology and Communication Sciences at Tampere University on 11 November 2026. The dissertation is in the field of mathematics.
The Opponent will be Associate Professor Przemysław Wałęga from the Queen Mary University of London, United Kingdom. The Custos will be Associate Professor Antti Kuusisto, Faculty of Information Technology and Communication Sciences.