Pierre-Emmanuel Jabin

Presentation - P.-E. Jabin joined the University of Maryland in 2011 and is currently the director of the Center for Scientific Computing and Mathematical Modeling. He defended his PhD in Mathematics in 2000 at the University Paris VI. He was an Assistant Professor at the Ecole Normale Superieure from 2000 till 2004, and a Professor at the University of Nice till 2011.
P-E Jabin works on transport and advection equations and multi-agent systems with applications to biology, ecology, statistical physics… He recently became interested in deep learning techniques
 He is an invited at the coming International Congress of Mathematicians in Rio de Janeiro.

Research project 

The research project proposes three examples connecting active areas of research to potential applications in data science:

- Multi-agent systems with synchronization mechanisms. Multi-agent systems are conceptually simple and exhibit a wide range of emerging macroscopic behaviors, making them useful in many different settings. We will consider systems where the connectivities between agents may be different and may evolve in time. This leads to learning systems where synchronization between agents is expected or desired, such as for the dynamics of neural networks.

- Feature extraction, for example for classification or clustering purposed, through transport operators. This point focuses on possible  extensions of the so-called Laplacian eigenmaps to  semi-supervised settings through advection operators. This type of methods is especially promising in clustering problems where a priori information on several clusters is available.

- Scaling and Stability of deep neural networks. The aim of the proposed work is to better understand Deep learning algorithms based on Artificial Neural Networks (ANN), and for example their stability during training through the connection between accuracy and the loss function.