Generalized Constraint Acquisition

摘要

Constraint programming is an approach to problem solving that relies on a combination of inference and search to solve real-world problems formulated as constraint satisfaction problems (CSPs). Many methods for solving CSPs have been developed. However, the specification of a CSP is sometimes not available, but may have to be learned from a training set, which is given, for instance, as a set of examples of its solutions and non-solutions. The motivating applications for constraint acquisition are many. For example, often one may wish to find a compact representation of a CSP instance for purposes such as explanation generation, requirements gathering, and specification. Acquiring soft constraints, which we focus on here, can be regarded as learning about preferences, uncertainty or costs in a combinatorial setting.