• Funding : Artois, Région HdF
  • Start year :
  • 2024

This thesis topic aims to create a bridge between artificial intelligence (AI) solutions and formal techniques of classical propositional logic. It seeks to develop generic methods for transforming AI models into propositional formulas, focusing on the satisfiability problem, known as the SAT. This approach looks to leverage the significant advancements made in the field of SAT solvers. A key aspect of this work is identifying the important features of AI models that can be efficiently translated into SAT representations. Particular interest is in translations that address fundamental issues, such as strengthening trust in algorithmic decisions, increasing transparency, and effectively integrating knowledge from the studied domains. The main goal is to establish an explanatory framework for AI models, which helps to detect potential biases and provides clear and accessible explanations to end-users.

More information on ADUM