Orbel Wolsey Award 2015


Broes De Cat graduated in 2009 with great honours as a civil engineer in computer science and A.I. from KU Leuven, Belgium. Shortly afterwards, he started a PhD under supervision of prof. Marc Denecker in the Declarative Languages and Artificial Intelligence group at KU Leuven. Supported by a grant from the Agency for Innovation by Science and Technology (IWT Vlaanderen), he investigated declarative approaches to search problems and integrated techniques from the fields of satisfiability checking, constraint programming, theorem proving and answer set programming into a state-of-the-art search algorithm. In May 2014, he successfully defended his thesis, titled “Separating Knowledge from Computation: an FO(.) Knowledge Base System and its Model Expansion Inference”.  He is currently working as a software engineer at OM Partners, a Belgian-based company developing and implementing leading supply chain planning software.


IDP + MinisatID – Multi-inference Knowledge Base System

A Knowledge Base System (KBS) aims to provide a user with a truly declarative language and a range of efficient inference engines, to allow to naturally express knowledge and to solve various tasks based on the modelled knowledge.

The IDP system is a such a KBS, based on the language FO(.), an extension of first-order logic that allows natural modelling of many common concepts, including definitional knowledge and aggregates. It provides for example inference engines to search for solutions to a set of constraints, to explain why no solution exists, to deduce whether a redundant, more efficient constraint is implied by the knowledge, etc. The solver MinisatID is the backend of several of those inference engines. It combines techniques from constraint programming with those from satisfiability checking (e.g., learning) and answer set programming (e.g., definitional knowledge handling) into a state-of-the-art combinatorial solver.

Broes latest research focused on improving the scalability of the inferences engines and the solver, by lazily decomposing constraints during search (instead of up-front as is the classical approach) and developing techniques to reduce the dependency of performance on how knowledge is modelled.

Print Friendly