guest ::
| Home > Publications database > Influence of Collective Network Dynamics on Stimulus Separation |
| Home > Publications database > Influence of Collective Network Dynamics on Stimulus Separation |
| Poster (After Call) | FZJ-2024-05812 |
; ; ; ; ;
2024
Abstract: Different stimuli elicit different transient neural responses in the brain. How is the information represented in the parallel neuronal activity of a neuronal population? We investigate this question in Neuropixel recordings of awake behaving mice by inferring the stimulus class (e.g., visual or tactile) from the activity. To quantify the separability, we use an optimally trained linear readout, which can be understood through the framework of Bayesian inference. We find that separability only depends on three geometric properties of the neuronal population vector, described by cones within the space of neuronal activity: The length $R$, the width of the cones $Q^{intra}$, given by the overlap of two vectors within classes, and the distance between two cones $Q^{inter}$, given by the overlap across two classes (A).We then analyze how the dynamics of a recurrent network deforms stimulus representations as a function of time. We employ a two-replica calculation to compute the time-evolutions of $R$, $Q^{intra}$ and $Q^{inter}$ In the limit of large networks, the dynamics of this model are fully described by these three quantities, which are matched to the corresponding experimental observables. The analytical theory, in turn, predicts the separability of the stimuli as a function of time and shows how the time-evolution of separability can be understood by the dynamic interplay between balance by inhibition, which controls the time course of $R$, and chaotic dynamics, which controls $Q^{intra}$ and $Q^{inter}$. Neither the decay of the firing rate alone nor that of the overlaps is predictive for separability, but rather their mutual relationships. For stimuli with reliable responses, the network can transiently increase the separability, as previously only seen for a constant population mean [1].To further probe the network's capacity for separation, we use mutual information to quantify the information contained in the population signal as a function of the number of stimuli. This reveals an optimal number of stimuli from the trade-off between encoding more information with more stimuli, and stimuli becoming less separable due to their increased overlap in neuron space (B). We show that the optimum depends on the population mean, allowing for the embedding of more stimuli for a smaller population mean.
|
The record appears in these collections: |
