Abstract

The dynamics in cortex is characterized by highly fluctuating activity: Even under the very same experimental conditions the activity typically does not reproduce on the level of individual spikes. Given this variability, how then does the brain realize its quasi-deterministic function? This variability has first been seen as a limitation for the concept of neural coding, but recent advances have questioned this traditional view and proposed possible functional roles for this variability. Here we propose a view where fluctuations are used as a basis to represent information. We show with the “covariance perceptron” how a fluctuation-based scheme can be powerful for the discrimination of spatio-temporal signals into several classes, and compare it with the traditional mean-based classification. Moreover, we explore the application to reservoir computing, where the information conveyed in the distinct statistical orders of time series is mixed, and test its performance on the discrimination of spoken digits. Together, this framework is a step toward biologically plausible learning rules for neuronal networks to extract complex statistical information from sensory signals, from principal component analysis to tensor factorization.

Ref:

Gilson M, Dahmen D, Moreno-Bote R, Insabato A, Helias M (2020) The covariance perceptron: A new paradigm for classification and processing of time series in recurrent neuronal networks. PLoS Comput Biol 16(10): e1008127. https://doi.org/10.1371/journal.pcbi.1008127

Join Zoom Meeting https://us02web.zoom.us/j/84616631650?pwd=LzRNL1MwcFI1cXZkdTZqNUI5NTIrQT09

Meeting ID: 846 1663 1650 Passcode: 072375