% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Schutzeichel:1044514,
author = {Schutzeichel, Lars and Bauer, Jan and Bouss, Peter and
Musall, Simon and Dahmen, David and Helias, Moritz},
title = {{D}ynamics of sensory stimulus representations in recurrent
neural networks and in mice},
school = {RWTH Aachen},
reportid = {FZJ-2025-03251},
year = {2025},
abstract = {Different stimuli elicit different transient neural
responses in the brain. But how is theinformation
represented in the parallel neuronal activity and how is it
reshaped by the dynamics oflocal recurrent circuits? We
investigate these questions in Neuropixels recordings of
awake behavingmice and recurrent neural network models by
inferring stimulus classes (e.g., visual or tactile) from
thenetwork activity. We derive a mean-field theory that
reduces the dynamics of complex networks to onlythree
relevant dynamical quantities that fully determine the
separability of stimulus representations.These dynamical
quantities geometrically represent the length of the neural
state vector, given by themean population activity $R$, and
the typical overlaps Q= and Q≠ of neural state vectors
representingthe variability of responses within or across
stimulus classes, respectively.Mean-field theory predicts
the time evolution of $R$, $Q^{=}$ , and $Q^{\neq}$ and
enables us to quantitatively explainexperimental
observables. The analytical theory predicts the temporal
dynamics of stimulus separabilityas an interplay of firing
rate dynamics and overlaps. It reveals how inhibitory
balancing controls the timecourse of $R$ and chaotic
dynamics control $Q^{=}$ and $Q^{\neq}$ , exposing the
mechanisms underlying separabilitybetween stimuli.The
analysis of mutual information of an optimally trained
readout on the population signal revealsa trade-off between
more information conveyed with an increasing number of
stimuli, and stimuli be-coming less separable due to their
increased overlap in the finite dimensional neuronal space.
We findthat the experimentally observed small population
activity $R$ is located in a regime where informationgrows
extensively with the number of stimuli, which is sharply
separated from a second regime, in whichinformation
converges to zero, revealing a crucial advantage of sparse
coding. Our work thus providesa novel understanding of
separability of stimuli shaped by collective network
dynamics.},
month = {Jul},
date = {2025-07-05},
organization = {CNS 34th Annual Computational
Neuroscience Meeting, Florence (Italy),
5 Jul 2025 - 9 Jul 2025},
subtyp = {After Call},
cin = {IBI-3 / IAS-6},
cid = {I:(DE-Juel1)IBI-3-20200312 / I:(DE-Juel1)IAS-6-20130828},
pnm = {5231 - Neuroscientific Foundations (POF4-523) / 5232 -
Computational Principles (POF4-523) / GRK 2416 - GRK 2416:
MultiSenses-MultiScales: Neue Ansätze zur Aufklärung
neuronaler multisensorischer Integration (368482240) / DFG
project G:(GEPRIS)533396241 - Evolutionäre Optimierung
neuronaler Netzwerkdynamik auf eine empfängerspezifische
interareale Kommunikation (533396241)},
pid = {G:(DE-HGF)POF4-5231 / G:(DE-HGF)POF4-5232 /
G:(GEPRIS)368482240 / G:(GEPRIS)533396241},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/1044514},
}