 |
Center for Nonlinear Studies |
The Integrate-and-Fire (IF) model, introduced more than one century ago, stands out among the highly non-linear dynamical models. Despite its simplicity, the IF model captures the essential features of a neural cell, and, as such, it has been widely used for decades in computational neurobiology.
In this talk I will address two aspects of the learning problem with one or more IF neurons:
1. to what extent can an IF neuron learn dynamical patterns of activities? This question is of importance to the highly debated question of the existence of time-based vs. frequency-based neural codes.
2. how can one infer the synaptic connections of an assembly of IF neurons from the observation of their spiking activity? A practical procedure will be proposed, and tested on synthetic data as well as multi-electrode recordings of retinal ganglion cells.
Host: Lenka Zdeborova