Yiannis Siglidis
10 Octobre 2019, 14:00. Salle 26-00 332, Jussieu.
The study of dynamical graphs, i.e. a data-model capable of modelling temporal-structures, becomes more and more important nowadays as the amount of data produced and the storing capabilities of the existing computational infrastructure associates information more and more with time. In their paper « Stream Graphs and Link Streams for the Modeling of Interactions over Time » of 2017, Matthieu Latapy, Tiphaine Viard, Clémence Magnien, defined a theoretical framework for modelling temporal-networks, that attempts a consistent temporal-generalisation of graph-theory. Under the founding of the ODYCCEUS(https://www.odycceus.eu/) interdisciplinary european program, a first attempt for implementing a library for stream-graphs, by Yiannis Siglidis (developer) and Robin Lamarche- Perrin (supervisor). Their implementation stands as a prototype of a generic basis that would support all existing implementations of stream-graph algorithms, while allowing enrichment and re-design of elementary data-structures. Its contribution shouldn’t only be considered as a technical one, in the sense of producing a new baseline-tool, but also as a theoretical one, as we examined and uncovered some of the immaturities of current from the view of implementation. We hope that this tool will show a small part of the computational possibilities that are provided by the formalism of stream-graphs and as so raise the importance of theoretical research in that direction as well as theoretical research around data-structures, supporting this novel approach on dynamical networks. Their current implementation supports continuous, discrete, instantaneous, weighted and unweighted instances of directed stream-graphs. The libary can be found at https://github.com/ysig/stream_graph and is licensed under GNU-GPL3.