Community Detection and Mean Field Approximation for Dimension Reduction of Spiking Network Models

Internal Seminar:  Carlos Gutierrez, Postdoctoral Researcher, Neural Computation Unit (Kenji Doya)

Title : Community Detection and Mean Field Approximation for Dimension Reduction of Spiking Network Models

Abstract :  Advances in serial section electron microscopy and optical imaging are making cellular connectomic data of local circuits available. Spiking network models based on connectomic data can help us understand how dynamic neuron-to-neuron interactions realize certain representations or computations in local circuits. However, when we try to understand the whole brain function, spiking network models can be impractical because of their huge computational requirements and a large number of parameters to be inferred. We propose a framework for network model reduction by combining a community detection algorithm and mean field approximation. We ran simulations of the original spiking network model and the reduced model, comparing mean firing rates under different external inputs. The proposed approach enables efficient simulation of a large-scale network model constrained by neuron-level connectomics data and can be used as a predictive tool for investigating the effects of different changes on neuronal properties and input stimuli.

Refreshments will be served afterwards.

We hope to see you there!

Sincerely,

Internal Seminar Organizing Committee:

Bianca Sieveritz

Rob Campbell

Maggi Mars-Brisbin

​ Yunhui Zheng

Jonathan Ward

Lauren Dembeck

Jigyasa Arora

Maéva Techer