Model Transformation as an Optimization Problem

Most of the available work on model transformation is based on the hypothesis that transformation rules exist and that the important issue is how to express them. But in real life, the rules may be difficult to define; this is often the case when the source and/or target formalisms are not widely used or proprietary. In this paper, we consider the transformation mechanism as a combinatorial optimization problem where the goal is to find a good transformation starting from a small set of available examples. Our approach, named model transformation as optimization by examples (MOTOE), combines transformation blocks extracted from examples to generate a target model. To that end, we use an adapted version of particle swarm optimization (PSO) where transformation solutions are modeled as particles that exchange transformation blocks to converge towards an optimal transformation solution. MOTOE has two main advantages: It proposes a transformation without the need to derive transformation rules first, and it can operate independently from the source and target metamodels.