A Preliminary Investigation into Automated Model Transformation Using Q-Learning: A Case Study

Model-Driven Engineering (MDE) emphasizes models as primary artifacts to enhance abstraction and automation in software development. However, manually defining model transformation rules is a time-consuming and error-prone task. This study proposes a Q-learning-based approach for automating model transformations by example, where transformation rules are dynamically learned through interaction with paired example models using reinforcement learning. The evaluation of the proposed method on UML-to-Java transformations demonstrates improvements in accuracy and scalability while reducing manual effort. The approach effectively addresses challenges such as model heterogeneities and dependency on high-quality examples, showcasing its potential to streamline and optimize model transformation processes.