Spontaneous emergence of fast attractor dynamics in a model of developing primary visual cortex

Abstract

Recent evidence suggests that neurons in primary sensory cortex arrange into competitive groups, representing stimuli by their joint activity rather than as independent feature analysers. A possible explanation for these results is that sensory cortex implements attractor dynamics, although this proposal remains controversial. Here we report that fast attractor dynamics emerge naturally in a computational model of a patch of primary visual cortex endowed with realistic plasticity (at both feedforward and lateral synapses) and mutual inhibition. When exposed to natural images (but not random pixels), the model spontaneously arranges into competitive groups of reciprocally connected, similarly tuned neurons, while developing realistic, orientation-selective receptive fields. Importantly, the same groups are observed in both stimulus-evoked and spontaneous (stimulus-absent) activity. The resulting network is inhibition-stabilized and exhibits fast, non-persistent attractor dynamics. Our results suggest that realistic plasticity, mutual inhibition and natural stimuli are jointly necessary and sufficient to generate attractor dynamics in primary sensory cortex.

Spiking attractor model of motor cortex explains modulation of neural and behavioral variability by prior target information

“…Over the last few decades, attractor dynamics, which is the mathematical instantiation of the hypothesis of cell assemblies (Hebb, 1949), has been established as the most viable computational model for this type of behavioral task. Yet, despite the fact that the basic theory of attractor dynamics is well understood (Hopfield, 1982;Amit & Brunel, 1997;Roudi & Latham, 2007) and there has been considerable experimental evidence to support it (Sakai & Miyashita, 1991;Wills et al, 2005;Knierim & Zhang, 2012;Miconi et al, 2016;Pereira & Brunel, 2018;Inagaki et al, 2019), the implementation of this type of framework in realistic spiking networks, and the link between the precise computational mechanisms and behavioral and neural trial-to-trial variability remain less clear. The classical cortical model is a balanced network of excitatory and inhibitory neurons (van Vreeswijk & Sompolinsky, 1998;Brunel, 2000) with random connectivity.…”

Cortical circuits implement optimal context integration

“…Bock et al [3] demonstrated, using optical physiology followed by electron microscopy (EM), a different rule for inhibitory neurons, with their inputs being nonspecific. This general principle of "like-to-like" connectivity when synaptic weights have reached their steady state is expected using a "cells which fire together, wire together" Hebbian principle [32,33,25].…”

Autonomous emergence of connectivity assemblies via spike triplet interactions

“…More generally, assemblies have been shown to emerge in recurrent network models with balanced excitation 471 and inhibition [47,48,117]. These assemblies exhibit attractor dynamics which have been argued to serve as 472 the substrate of different computations, such as predictive coding through the spontaneous retrieval of evoked 473 response patterns [47,48,117].…”

“…More generally, assemblies have been shown to emerge in recurrent network models with balanced excitation 471 and inhibition [47,48,117]. These assemblies exhibit attractor dynamics which have been argued to serve as 472 the substrate of different computations, such as predictive coding through the spontaneous retrieval of evoked 473 response patterns [47,48,117]. We investigated how structure emerges primarily from spontaneous dynamics in 474 on the fact that modularity of a network is closely related to the structure of the eigenvalue spectrum of its 656 weight matrix [94,122,123], high modularity means more strongly embedded communities.…”