Page 22 - Rappaport Institute Magazine 2024

BRAIN SCIENCE
Hadas Benisty, PhD
Senior Lecturer of Computational Neuroscientist
Interpretable Dynamic Models for Network Dynamics in Learning Animals
My research is primarily focused on the dynamics of neuronal networks and how their organizational principles, functional connectivity and plasticity mechanisms relate to behavior and learning.
We propose a paradigm shift from conventional statistical analysis of functional connectivity as a static entity, to a geometrical point of view of connectivity as a high-dimensional and dynamic process. In our collaboration with Jackie Schiller and Ronen Talmon we used Graph-Theory and Riemannian Geometry to model the temporal transformation of cortical networks in the primary motor cortex (M1) during learning of a motor task. We show that 1) while the average activity and connectivity remain stable, the activity kinetics and the correlational configuration gradually transform towards an expert configuration and 2) dopaminergic activation in M1 is essential for this process to transpire.
To further investigate the dynamic mechanisms of this transformation we collaborate with Gal Mishne, Ofir Lindenbaum and Simon Musall to develop interpretable deep learning architectures. Using this approach, we detect sub-networks emerging with learning which encode specific behaviors, as a function of context such as time or sensory stimuli.
Selected Publications
ˆ Ganayim* A., Benisty* H., Cohen-Rimon A., Schwartz S., Talmon R., and Schiller (2023) J. VTA projections to M1 are essential for reorganization of layer 2-3 network dynamics underlying motor learning
Sristi R., Lindenbaum O., Lavzin M., Schiller J., Mishne G. and Benisty H. (2023). Contextual Feature Selection with Conditional Stochastic Gates
ˆ Benisty, H.*, Barson, D.*, Moberly, A., Lohani, S., Coifman, R. R., Mishne, G., Cardin, J. A. and Higley, M. J. (2023). Rapid fluctuations in functional connectivity of cortical networks encode spontaneous behavior. Nature Neuroscience
Collaborators
• Simon Musall – Research Center Jülich • Shreya Saxena, Yale University
• Gal Mishne – UCSD
• Ofir Lindenbaum – Bar Ilan University
hadasbe@technion.ac.il
Hadas Benisty Lab
PhD, 2005 – Technion, Israel
Top: Experimental design for learning a motor task for the control and manipulated mice groups during VTA silencing with local CNO in M1. Middle - schematic illustration of the analysis pipeline. Construction of correlation matrices of
the activity between pairs of neurons, used to evaluate
the Riemannian centroid representing the functional connectivity of the network in a given training session. Bottom: Example of a two-dimensional embedding of trial-based correlation matrices (colors indicate training sessions) and their Riemannian centroids (black) for a control (left) mouse and a manipulated mouse (right).