Community Research and Development Information Service - CORDIS

ERC

Predispike Report Summary

Project ID: 312227
Funded under: FP7-IDEAS-ERC
Country: France

Mid-Term Report Summary - PREDISPIKE (Spike-based predictive coding: Closing the loop between neural dynamics and computation)

Brain circuits are exquisitly organised and in particular, it is clear that inhibition and inhibitory interneurons play a crucial role. Contrary to initial beliefs, it appears that most stimuli or action excite and inhibit a neuron simultaneously and almost equally. This is called excitatory/inhibitory balance, and we have been exploring the role and implication of this balance for neural coding, computation and behavior. We found that balanced networks exhibit large variability, but that this variability is not noise, but degenaracy. As many different patterns of neural responses code for the same information, networks switch between them effortlessly (these are all the states for which the network is balanced), making single neurons look noisy. But in those networks, every single spike counts, and the global representation is much more precise than is currently thought. This predicts low dimensional structures in neural activity, that we are looking for in data recorded from mouse and zebrafish. In parallel, we explored the implication of E/I balance at a much larger scale, i.e. in the representation we construct in our hierarchical brain. Because the brain is highly recurrent, it is full of loops where information flows back and forth uncontrollably. Seeing green could make believe in trees and then believe in green...etc. This problem can be avoided by E/I balance because every redundant (predictable) information is cancelled automatically. Our hypothesis is that this balance can be broken in some pathology like schizophrenia or autism, leading to the formation of abberent beliefs and percepts. We have validated this theory with well chosen behavioral tasks.

Contact

Damien VOGEL
Tel.: +33 1 44 32 29 23
Fax: +33 1 44 32 29 33
E-mail
Record Number: 187794 / Last updated on: 2016-08-24