DiverSE Coffee [Hugo Martin]: When configurable systems meet Machine learning

Abstract

Modern configurable systems are designed in a powerful and flexible way in orderto meet the needs of more and more users at the expense of their clarity anduser-friendliness. With huge configuration space, it is impossible to fullyexplore strong interactions between parameters that even experts struggle tofind a configuration matching their needs. In order to help users to quicklyfind a configuration corresponding to their expectations, we propose anautomated specialization process based on machine learning. By studying twodifferent approaches and many aspects of the learning while testing the processon many real-world systems, we came to the conclusion that such process offersgood results as well as it opens many ways of improvement.

Date
Event
DiverSE coffee
Location
Rennes, France
Modern configurable systems are designed in a powerful and flexible way in order to meet the needs of more and more users at the expense of their clarity and user-friendliness. With huge configuration space, it is impossible to fully explore strong interactions between parameters that even experts struggle to find a configuration matching their needs. In order to help users to quickly find a configuration corresponding to their expectations, we propose an automated specialization process based on machine learning. By studying two different approaches and many aspects of the learning while testing the process on many real-world systems, we came to the conclusion that such process offers good results as well as it opens many ways of improvement.