Towards Model-Based Optimisation: Using Domain Knowledge Explicitly

TL;DR: A first prototype for running search algorithms directly on models themselves (rather than a separate representation) is presented and key research challenges for this approach to model optimisation are derived.

Abstract: Search-based software engineering (SBSE) treats software-design problems as search and optimisation problems addressing them by applying automated search and optimisation algorithms. A key concern is the adequate capture and representation of the structure of design problems. Model-driven engineering (MDE) has a strong focus on domain-specific languages (DSLs) which are defined through meta-models, capturing the structure and constraints of a particular domain. There is, thus, a clear argument for combining both techniques to obtain the best of both worlds. Some authors have proposed a number of approaches in recent years, but these have mainly focused on the optimisation of transformations or on the identification of good generic encodings of models for search. In this paper, we first provide a structured overview of the current state of the art before identifying limitations of the key proposals (transformation optimisation and generic genetic encodings of models). We then present a first prototype for running search algorithms directly on models themselves (rather than a separate representation) and derive key research challenges for this approach to model optimisation.