Identifying system dynamics with machine learning.
In this meetup, we introduce machine learning techniques dedicated to the identification of system dynamics. From discrete observations of the system, we determine the local influences of its components. In the method we develop, the observations we consider as input are the state transitions of the system. From these observations, we build and refine, transitions after transition, a logic program that captures the dynamics of the system. This method allows, among other things, to learn Boolean networks and to identify cellular automata. This technique can be applied to bioinformatics, especially for the identification of a gene regulatory network from observations obtained through laboratory experiments.
Tony Ribeiro did a PhD at the National Institutes of Informatics (NII) of Tokyo, focusing on the development of methods for learning system dynamics. During his postdoc at Laboratoire des Sciences du Numérique de Nantes (LS2N) he challenged his machine learning techniques on real data coming from biology. His main goal is the production of predictive model. Taking time series data as input: we have to find the genes level of expressions, capture the variables interactions and finally produces predictions under unobserved conditions.
- Katsumi Inoue, Tony Ribeiro, Chiaki Sakama: Learning from interpretation transition. Machine Learning 94(1): 51-79 (2014). http://www.wakayama-u.ac.jp/~sakama/papers/MLJ2013.pdf
- Tony Ribeiro, Katsumi Inoue: Learning Prime Implicant Conditions from Interpretation Transition. ILP 2014: 108-125. https://dtai.cs.kuleuven.be/events/ilp2014/submissions/ilp2014_submission_9.pdf
- Tony Ribeiro, Sophie Tourret, Maxime Folschette, Morgan Magnin, Domenico Borzacchiello, Francisco Chinesta, Olivier Roux, Katsumi Inoue. Inductive Learning from State Transitions over Continuous Domains, The 27th International Conference on Inductive Logic Programming, (ILP 2017), Orléans, France. https://hal.archives-ouvertes.fr/hal-01655644/document
La présentation sera en français.