Constraint Programming on Infinite Data Streams

Arnaud Lallouet (1) , Yat Chiu Law (2) , Jimmy H.M. Lee (2) , Charles F.K. Siu (2)

1 Equipe CODAG - Laboratoire GREYC - UMR6072
2 CSE - Department of Computer Science and Engineering [HKUST]

Classical constraint satisfaction problems (CSPs) are commonly defined on finite domains. In real life, constrained variables can evolve over time. A variable can actually take an infinite sequence of values over discrete time points. In this paper, we propose constraint programming on infinite data streams, which provides a natural way to model constrained time-varying problems. In our framework, variable domains are specified by ω-regular languages. We introduce special stream operators as basis to form stream expressions and constraints. Stream CSPs have infinite search space. We propose a search procedure that can recognize and avoid infinite search over duplicate search space. The solution set of a stream CSP can be represented by a B¨uchi automaton allowing stream values to be non-periodic. Consistency notions are defined to reduce the search space early. We illustrate the feasibility of the framework by examples and experiments.

Domaines: Intelligence artificielle [cs.AI]

Arnaud Lallouet, Yat Chiu Law, Jimmy H.M. Lee, Charles F.K. Siu. Constraint Programming on Infinite Data Streams. International Joint Conference on Artificial Intelligence, Jul 2011, Barcelone, Spain. pp.597-604. ⟨hal-01009693⟩