Towards Explanable AI by Reasoning with (Smart) Constraints •
- Directeur de thèse :
- Christophe Lecoutre
- Financement : Artois
Constraint Programming (CP) is a general framework providing efficient models and algorithms for solving combinatorial constrained problems. The research conducted at CRIL (AI Laboratory situated at Lens, France) about CP aims at developping generic approaches (i.e., AI-related) for both CSP/COP (Constraint Satisfaction/Optimization Problem) and SAT (Satisfiability Testing). Useful tools for conducting innovative research at CRIL include state-of-the-art academic solvers (the CSP solver AbsCon and the SAT solver Glucose), the integrated representation format XCSP3, the modeling Python library PyCSP3 (and the modeling Java API JvCSP3), as well as a computing cluster.
The thesis is about studying how (new extended forms of) so-called smart constraints can be useful for efficiently representing large parts of models and contributing to the challenging issue of explainable AI, as explained below.