Digital Continuity Based on Reinforcement Learning Model Transformation

Abstract

With the importance gained by Service-Oriented Architectures (SOA) to simplify and decompose complex enterprise information system into autonomous, modular, reusable and, flexible model, the need to make models interoperable to ensure digital continuity has increased. However, the structural, syntactic, and semantic heterogeneity of metamodel, drastically complicates the interconnection of models and thus the digital continuity. A key element of Model-Driven Architecture (MDA), model transformations could be one of the solutions to promote model interoperability. They allow the description of the transformation rules that link the different metamodels concepts. However, significant efforts are needed to describe and maintain model transformations in an ever-changing digital environment. One of the key challenges of the MDA approach is the automation of model transformations. Recent work exploiting machine learning techniques to infer model transformations has shown very promising results. However, learning algorithms, involve too much training data and require a large and varied dataset. In our work, we want to experiment the reinforcement learning techniques to infer transformation rules and to counter the need to provide a considerable volume of data for machine learning.