Cendra Agulhon



Salle des Thèses -Jacob Building- 5th floor


16 Jan 2023


11 h 00 min - 12 h 00 min


INCC Seminar Series

From structure and dynamics of biological networks to spiking neural networks and generalized artificial intelligence, by Samuel Bottani

From structure and dynamics of biological networks to spiking neural networks and generalized artificial intelligence
In this talk I will first present the line of work that guided my activity between physics and biology. The broad topic concerns the structure and dynamics of biological networks, ranging from system biology to neuronal networks. Biological activity at different scales is indeed often the emerging result of the topology of possible interactions between the involved parts combined with non-linear properties of reactions. In this frame, my studies concerned gene expression dynamics, control of genetic networks, effects of gene copy numbers and lately the spatio-temporal activity in neural cultures. I have been particularly interested in studying ignition properties of large scale neural activity in neural populations depending on the underlying network connectivity, properties of neural networks’ bursts, as well as modeling of neurons’ and networks’ morphogenesis in structured micro environments.
In the second part of this talk I will present new topics I am getting involved in e and for which I am actively looking for collaborations and experimental settings to work with. As an evolution from the emergent neural structure and dynamics I was previously studying I am currently exploring “Spiking Neural Networks” (SNN) based approaches in Artificial Intelligence. Inspired by computational neurosciences and neuromorphic computation these techniques differ from the formal neural networks of “classical AI” and deep learning because of their asynchronous temporal mode of operation and the coding of information in the form of temporal sequences of action potentials.
The lack of appropriate hardware, the non-availability of adapted learning algorithms, at the same time as the fantastic progress and success of classical machine and deep learning have slowed down the development of SNN based techniques. However, massively parallel electronic implementation of units simulating spiking neurons as well as of interactions reproducing the time-dependent synaptic plasticity of biology has also progressed. This makes it now possible to consider practical applications of embedded systems based on SNNs with very low energy consumption.
Additionally, SNNs naturally possess progressive learning capabilities that adapt over time, which are interesting for applications with unexpected signals, relevant for example in robotics, BCI applications or exploration of cognitive paradigms.

Short Biography
Samuel Bottani
“Trained as a physicist I’m fascinated by viewing Life at all scales through the eyes of mathematical and physical modeling. Associate Professor at the Université Paris Diderot Physics Department in the “Matières et Systèmes Complexes” lab and theory group, I was recruited as Full Professor at the Université Paris Descartes, Fundamental and Biomedical Sciences Department in 2019, joining the MSCmed sub-unit.
Besides the science presented in the seminar, I am highly interested in pedagogy for higher education, interdisciplinary and project based teaching, and usage of technology for teaching and learning. Strongly committed in my academic environment, I was especially involved in the development and management of doctoral education programs as doctoral school director (ED Frontières du Vivant) and director of the Center for Doctoral Training in Professional Development. Currently I’m co-head of the Bioengineeging and Innovation in Neurosciences track of the Biomedical Engineering Master program of Université Paris Cité, ENSAM and PSL, and co-director of the Graduate School of Biomedical Engineering. Facing the climate crisis I am more and more concerned about adaptation and sustainability in higher education institutions.”

Web Site