A three-year PhD position is available at the University of Lille, France in collaboration with HUN-REN Institute for Computer Science and Control, Hungary.
Topic: Theoretical foundations of learning robust neural ODEs
Job description: Neural Ordinary Differential Equations (neural ODEs) are dynamical models blending neural networks and dynamical systems. They have gained popularity in the recent years mainly because of their attractive properties, e.g., ability to take into account physical constraints, to model irregularly sampled time-series, provide classifiers which are robust to adversarial attacks, predict long range dependencies, ability to serve as generative models (images, etc.). Despite many advances, there are important gaps in learning theory of such systems. One such gap is lack of formal guarantees for the generalization error, i.e., the consistency of the learning algorithm.
In this thesis we aim at filling this gap by formalizing neural ODEs and providing theoretical guarantees for learning robust neural ODEs. The methodology relies on combining ideas from control theory on stability and robustness of dynamical systems and their learning with approaches from statistical machine learning.
More details can be found at
https://phd.cristal.univ-lille.fr/details.html?id=86801c70f7154f65bff89bb67b97c3cd
If you are interested, please send the following documents to Mihaly Petreczky (mihaly.petreczky@centralelille.fr), Andras Benczur (benczur@ilab.sztaki.hu), Daroczy Balint (daroczy.balint@sztaki.hu), Ying Tang (ying.tang@univ-lille.fr).
-Detailed CV
-Motivation letter
-Transcripts of records (Undergraduate degree and Master degree)
-Recommendation letters
-Research-oriented documents (e.g. thesis, conference/journal publication)
