High-level description. We study structural, coding and dynamical properties of biologically plausible spiking networks implementing efficient coding. A network with optimal parameters generates biologically plausible spiking dynamics. On the coding level, the model implements lateral inhibition among excitatory neurons with similar selectivity. Lateral inhibition relies on structured E-I-E connectivity that is optimal as per analytical calculations, given the assumption on the computation performed by the network. Structured connectivity in turn predicts specific levels of the instantaneous E-I balance that is necessary to support lateral inhibition. Optimal parameters of such network match biophysical parameters typically observed in biological networks in sensory cortex.
Paper
Conference talks
Koren V, Malerba S, Schwalger T, Panzeri S (2023), Efficient computing of high-dimensional representations with biologically plausible spiking neural networks. Spiking Networks as Universal Function Approximators (SNUFA) conference 2023 (online); YouTube link
Koren V, Schwalger T (2021) Computation through neural dynamics of a spiking network with predictive coding. Computational Neuroscience Meeting (OCNS), online.
Conference posters
Koren V, Blanco-Malerba S, Schwalger S, Panzeri S (2024), Structure, dynamics, coding and optimal biophysical parameters of efficient excitatory-inhibitory spiking networks. 5th International Convention on Mathematics of Neuroscience and AI, Rome, Italy
poster_roma_2024.pdf
Koren V, Schwalger T, (2021) Functional E-I spiking network with predictive coding. Computational and Systems Neuroscience conference (CoSyNe)