With the target of providing an enhanced toolkit for designing complex systems by combining logic and intuition, we will extend the theory, algorithms and implementation underlying our guaranteed graphical model solver (toulbar2) to address more complex reasoning problems for NP-hard problem solving from optimization, quantified reasoning or counting.
To guide reasoning so that it both finds solutions faster and is able to take into account information extracted from data by Machine Learning, we will integrate ML technology inside our algorithms and models to also solve multi-criteria problems that account both for physical laws represented as logical rules or criteria as well as ML extracted information.
The design of complex molecular systems such as proteins will be used as our main target throughout the project, to validate and enhance the visibility of our progresses.
Programs : IA acceptable, certifiable & collaborative
Thèmes : apprentissages avec peu de données ou des données complexes, IA et modèles physiques, Raisonnement automatique et décision, explicabilité
Thomas Schiex, DR Inra, MIAT
Chercheurs associés :
Sophie Barbe, INSA Toulouse,
David Simoncini, Irit