A DSL for Encoding Models for Graph-Learning Processes

Zahra Rajaei (1) , Shekoufeh Kolahdouz-Rahimi (1) , Massimo Tisi (2, 3) , Frédéric Jouault (4)

1 University of Isfahan
2 IMT Atlantique - DAPI - Département Automatique, Productique et Informatique
3 LS2N - équipe NaoMod - NaoMod - Nantes Software Modeling Group
4 ÉRIS - ESEO-ÉRIS

Specific deep-learning tools for graph-structured data, i.e. graph-learning, are successfully used in several domains. Their use in Model-Driven Engineering (MDE) requires MDE practitioners to have a good understanding of technical aspects of the graph-learning process. For instance, automatic translators need to be developed, in order to encode models in the most effective input format for deep-learning neural networks. With this work, we aim at assisting MDE practitioners in applying deep learning on their models. For this purpose, we introduce a Domain-Specific Language (DSL) for configuring the encoding of models into suitable input for graph-learning tools. This DSL is interpreted to automatically translate MDE datasets, enabling their use in machine-learning pipelines. To evaluate this research, we consider the AIDS dataset as instances of a corresponding metamodel. We use our DSL to automatically encode models of this dataset into the format expected by a graph-learning tool. The experimental evaluation demonstrates that we are able to obtain the same encoding used in related work.

Domaines: Informatique [cs]

Zahra Rajaei, Shekoufeh Kolahdouz-Rahimi, Massimo Tisi, Frédéric Jouault. A DSL for Encoding Models for Graph-Learning Processes. 20th International Workshop on OCL and Textual Modeling, Jun 2021, Bergen, Norway. ⟨hal-03252919⟩