Integrating the Support for Machine Learning of Inter-Model Relations in Model Views

By: James Pontes Miranda, Hugo Bruneliere, Massimo Tisi, Gerson Sunyé

Abstract

Model-driven engineering (MDE) supports the engineering of complex systems via multiple models representing various aspects of the system. These interrelated models are usually heterogeneous and specified using complementary modeling languages. Thus, model-view solutions can be employed to federate these models more transparently. Inter-model links in model views can sometimes be automatically computed via explicitly written matching rules. However, in some cases, matching rules would be too complex (or even impossible) to write, but inter-model links may be inferred by analyzing previous examples instead. In this paper, we propose a Machine Learning (ML)-backed approach for expressing and computing such model views. Notably, we aim at making the use of ML in this context as simple as possible. To this end, we refined and extended the ViewPoint Definition Language (VPDL) from the EMF Views model-view solution to integrate the use of dedicated Heterogeneous Graph Neural Networks (HGNNs). These view-specific HGNNs are trained with appropriate sets of contributing models before being used for inferring links to be added to the views. We validated our approach by implementing a prototype combining EMF Views with PyEcore and PyTorch Geometric. Our experiments show promising results regarding the ease-of-use of our approach and the relevance of the inferred inter-model links.

Keywords

MDE, Modeling languages, Model Views, Machine Learning, Graph Neural Networks

Cite as:

James Pontes Miranda, Hugo Bruneliere, Massimo Tisi, Gerson Sunyé, “Integrating the Support for Machine Learning of Inter-Model Relations in Model Views”, Journal of Object Technology, Volume 23, no. 3 (July 2024), pp. 1-14, doi:10.5381/jot.2024.23.3.a4.

PDF | DOI | BiBTeX | Tweet this | Post to CiteULike | Share on LinkedIn

The JOT Journal   |   ISSN 1660-1769   |   DOI 10.5381/jot   |   AITO   |   Open Access   |    Contact