ParXCel - Machine Learning and Parallelization for Scalable Constraint Solving

In ParXCel, we will focus on the development of synthesis (e.g., configuration) and analysis algorithms (e.g., conflict detection and diagnosis) that help to tackle the above mentioned challenges. The overall idea of ParXCel is 1) to develop machine learning techniques for boosting the performance and prediction quality of constraint solving and 2) to develop parallelized approaches for efficient analysis operations (conflict detection and diagnosis). Major research contributions of ParXCel will be the following. First, we will focus on the integration of machine learning techniques with adaptive search heuristics. This will make it possible to transfer machine learning based prediction approaches to configuration scenarios and optimize the prediction of future user preferences. Second, we will parallelize existing algorithmic approaches especially in conflict detection and direct diagnosis [JUN2004, FSZ2012]. Parallelization on the algorithmic level, for example, on the level of conflict detection, provides the possibility of exploiting environments such as Java ForkJoin that support the implementation of parallelized algorithms.