Here are presented the different research project on which I have worked since I started my PhD. I am interested in machine learning processes, especially Relational Concept Analysis which I used several times, and most of my work turns around the idea of automatically finding semantical relations between different kind of artifact that do not appear to be linked a priori. At first in the software engineering domain to learn model transformations, and to provide interoperability between components during my postdoc. Then in the data mining domain to find implication rules between several sets of relational data.
The work of my postdoc is focused on the elaboration of a new data mining approach based on Relational Concept Analysis and applied on hydroecological data from the FRESQUEAU project. Our goal is to be able to assert relations between faunistic characteristics of a watercourse and its physico-chemical characters. The challenge here is to be able to handle relational data at a large scale.
As part of this work is developped RCAExplore, a new implementation of the RCA process allowing more customisation of the process in order to explore dynamically different configurations for building relational concepts.
There exist other methods besides RCA to handle relational data. Comparing these other methods can help us to understand RCA process and can bring new features to our methods.
The work of my postdoc was in the context of the ITEA project OPEES. The goal was to provide methods to ease composability of OPEES components. We used ModMap to implement an alignment between SmartQVT and Kermeta in order to be able to call SmartQVT transformation in a kermeta environment.
My different works during my thesis can be separated in different parts that led to the creation of a tool for validation purpose
(By-Example RCA MOdel Transformation Elaboration)
My main topic of research is the generation of model transformations based on examples using RCA, a classification process based on concept lattice theory.(more details and a tool coming soon)
One particularity of most of the existing MTBE techniques is the need of an alignment between models elements. For an early version of the tool: see here (more details coming soon)
(Enhancing Usecases using RElational Concept Analysis)
In software engineering, usecases are models of the available interactions between a system and other actors. Those models are usually considered as part of requirements and there exists many languages to describe them. The usecases on which we focus here are UML use cases. EURECA is a system restructuring a UML use case using Formal and Relational Concept analysis in order to obtain refactored models easier to read.