Date: October 04, 2016. 11:00
Location: CCU Seminar Room
Title: Spiking neurons can discover predictive features by aggregate-label learning
Affiliation: Max Planck (Experimental Medicine), Goettingen.
The Max Planck Research Group Theoretical Neuroscience uses analytical and numerical modeling techniques to identify the computational principles underlying spike based information processing and learning in central nervous systems and to understand how these principles are implemented by biological processes. Specifically, we focus on the role of action potential timing in subserving sensory neuronal representations and computation as well as in controlling synaptic plasticity. By advancing the recently developed tempotron family of spiking neuronal network models (Gütig & Sompolinsky, 2006) to, firstly, incorporate biophysically realistic cellular and synaptic processes, and secondly, boost their applicability to multi-layer neuronal networks, we strive for a qualitatively new functional level of neuronal systems modeling. Being able to subsume different levels of abstraction within a given computational perspective, this approach will bridge the gaps between cellular biophysics, mechanisms of synaptic plasticity and neural systems function. By engaging spiking neural network models in functional contexts of biologically realistic sensory information processing tasks, we will proceed along three interrelated lines of research
Single cell computation: What are the functional roles of specific cellular processes, such as conductance kinetics, short-term synaptic dynamics or active dendritic compartments in spike based neural processing?
Neuronal coding: What coding and readout strategies underlie spike timing based neuronal representations and their transformations across successive sensory processing stages?
Distributed multi-layer neuronal processing and learning: What are the computational principles underlying spike based learning rules for synaptic efficacies and their short-term dynamics in multi-layer neural networks engaged in demanding sensory processing tasks?
Each of these lines of research interfaces naturally with parallel experimental approaches that probe the nervous system at different levels of resolution.