Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : LS2N
Durée : 5 à 6 mois
Contact : julien.blanchard@univ-nantes.fr
Date limite de publication : 2024-04-20

Contexte :
With the digital transformation of our societies, more and more domains are interested in the analysis of series of symbolic events, such as Web user behavior analysis, healthcare and patient monitoring or fraud detection in financial transactions. Grasping the dynamics of these data streams requires revealing hidden structures behind them. To this end, Gunawardana and Meek have proposed a novel approach called Graphical Event Models (GEMs), probabilistic models capable of explicitly describing the temporal dependencies between events [1,2,3]. The PILGRIM software library developped by the DUKe reasearch team at the LS2N laboratory provides an original implementation of GEMs and of algorithms for learning them from logs – see [4,5,6] for examples of DUKe’s works on GEMs.

Sujet :
The goal of this internship is to study GEMs more deeply from both a theoretical and practical point of view, and to develop new methods for learning and using them. Several directions are possible:
– design and implement sampling and inference algorithms ;
– compare GEMs with process mining (log analysis approach [7]) : founding principles, model properties, experiments on data ;
– adapt explainable AI techniques to GEMs.

Possibility of extending into a PhD thesis if the internship is satisfactory.

[1] Asela Gunawardana and Chris Meek. Universal models of multivariate temporal point processes. In Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, volume 51 of Proceedings of Machine Learning Research, pages 556–563, 09–11 May 2016. PMLR.

[2] Christopher Meek. Toward learning graphical and causal process models. In Proceedings of the UAI 2014 Conference on Causal Inference: Learning and Prediction – Volume 1274, CI’14, pages 43–48, 2014. CEUR-WS.org.

[3] Asela Gunawardana, Christopher Meek, and Puyang Xu. A model for temporal dependencies in event streams. In Proceedings of the 24th International Conference on Neural Information Processing Systems, NIPS’11, pages 1962–1970, 2011. Curran Associates Inc.

[4] Mathilde Monvoisin and Philippe Leray. Multi-task transfer learning for timescale graphical event models. In 15th European Conference on Symbolic and Quantitative Approaches to Reasoning with Uncertainty (ECSQARU 2019), 2019.

[5] Dimitri Antakly, Benoit Delahaye, and Philippe Leray. Graphical event model learning and verification for security assessment. In 32th International Conference on Industrial, Engineering, Other Applications of Applied Intelligent Systems (IEA/AIE 2019), Advances and Trends in Artificial Intelligence, pages 245–252, 2019.

[6] SPARS project : Sequential pattern analysis in robotic surgery. https://project.inria.fr/spars/

[7] Wil M. P. van der Aalst. Process Mining: Data Science in Action. Springer, Heidelberg,2 edition, 2016.

Profil du candidat :
– Autonomy, rigor, initiative, analytical and synthesis skills, sense of commitment, communication and interpersonal skills
– C++ programming

Send CV, motivation letter and academic results in PDF format to julien.blanchard@univ-nantes.fr

Formation et compétences requises :
Master’s degree with specialization in Data Science or Machine Learning
Master 2 ou dernière année d’école d’ingénieur

Adresse d’emploi :
Polytech Nantes, rue Christian Pauc, 44306 Nantes, France

Document attaché : 202311301055_Internship_GEM_2024.pdf