Deep learning representations and strategies for the identification of dynamical systems
Supervisors: Ronan Fablet, Cédric Herzet (firstname.lastname@example.org)
Research team: IMT Atlantique, Lab-STICC, TOMS, Brest
Expected duration: 6 months
Deep learning  has experienced tremendous growth in a few years in the field of artificial intelligence and computer vision. Initially exploited for classification and recognition problems, it has also become a reference framework for the resolution of signal and image processing problems: image synthesis, super-resolution, denoising, inpainting, segmentation ....
Data-driven strategies for the analysis, modeling and reconstruction of dynamical systems are also currently emerging as promising research directions as an alternative to classic model-driven approaches for a wide variety of application fields, including atmosphere and ocean science, applied physics,.... [2,3,4]. In this context, this internship will aim to explore deep learning strategies and neural network (NN) representations for the identification of dynamical systems (i.e., the differential equation governing a dynamical system) . Both learning-from-data and adversarial-learning self-learning strategies will be of interest [2,5].
Applications to both low-dimensional chaotic systems and spatio-temporal fields (including images) will be explored depending on the progress of the work. All experiments will be implemented under Python using dedicated libraries such as Keras and/or Tensorflow frameworks.
Keywords: neural networks, learning strategies, inverse problems, dynamical systems, model identification, application to image time series
The envisioned workplan involves three main aspects:
 Y. LeCun, Y. Bengio, and G. Hinton. Deep learning. Nature, 521(7553) :436– 444, May 2015.
 R. Lguensat, P. Tandeo, P. Aillot, R. Fablet. The Analog Data Assimilation. Monthly Weather Review, 2017.
 R. Lguensat, P. Viet, M. Sun, G. Chen, F. Tenglin, B. Chapron, R. Fablet. Data-driven Interpolation of Sea Level Anomalies using Analog Data Assimilation. Submitted
 R. Fablet, P. Viet, R. Lguensat. Data-driven Methods for Spatio-Temporal Interpolation of Sea Surface Temperature Images. IEEE Trans. on Computational Imaging, 2017.
 A. Radford, L. Metz, S. Chintala. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. https://arxiv.org/abs/1511.06434
 R. Fablet, S. Ouala, C. Herzet. Bilinear residual Neural Network for the identification and forecasting of dynamical systems. Submitted
(c) GdR 720 ISIS - CNRS - 2011-2018.