Thomas Wennekers
Stochastic Interaction in Neural Systems
Wennekers, Thomas; Ay, Nihat; Andras, Peter
Authors
Nihat Ay
Prof Peter Andras P.Andras@napier.ac.uk
Dean of School of Computing Engineering and the Built Environment
Abstract
Brains are information processing systems evolutionary adapted since millions of years. It seems plausible therefore that they follow information theoretic optimization principles. In the early visual system mutual information maximization (Infomax) is one such widely accepted principle which claims that neurons represent as much information as possible in their output about the ensemble of stimuli at their input. The principle minimizes coding redundancy in feedforward pathways thereby efficiently coupling the brain to its environment, and it supports self-organization of fairly realistic receptive fields. Independent information channels seem useful for early stimulus coding under a bot-tleneck constraint, but brain activity is more complex in that cortical neurons get dynami-cally engaged into functional networks. They interact massively within and between areas in order to integrate distributed information into functional ensembles as tentatively ex-pressed by highly dynamic correlation patterns in their activity. Synchronized oscillations and synfire-chain-like spatio-temporal firing patterns are the two best known correlation patterns suggested to serve cortical integration. The maximization of spatio-temporal stochastic interactions (called TIM for temporal Infomax) has been proposed as an information-theoretic organizing principle in neural sys-tems which supports complex correlation patterns and a high cooperativity among cells. Instead of information transfer between input and output units it maximizes informa-tion exchange between all units in a system thereby emphasizing dynamic cooperation in recurrent networks in favor of static feedforward encoding systems only.
Citation
Wennekers, T., Ay, N., & Andras, P. (2005, August). Stochastic Interaction in Neural Systems. Presented at 6th Neural Coding Workshop, Marburg, Germany
Presentation Conference Type | Conference Paper (published) |
---|---|
Conference Name | 6th Neural Coding Workshop |
Start Date | Aug 23, 2005 |
End Date | Aug 28, 2005 |
Publication Date | 2005 |
Deposit Date | Nov 24, 2021 |
Book Title | 6th Neural Coding Workshop |
Public URL | http://researchrepository.napier.ac.uk/Output/2809053 |
You might also like
Structural Complexity and Performance of Support Vector Machines
(2022)
Presentation / Conference Contribution
Federated Learning for Short-term Residential Load Forecasting
(2022)
Journal Article