Recruitment and Consolidation of Cell Assemblies for Words by Way of Hebbian Learning and Competition in a Multi-Layer Neural Network

Abstract

Current cognitive theories postulate either localist representations of knowledge or fully overlapping, distributed ones. We use a connectionist model that closely replicates known anatomical properties of the cerebral cortex and neurophysiological principles to show that Hebbian learning in a multi-layer neural network leads to memory traces (cell assemblies) that are both distributed and anatomically distinct. Taking the example of word learning based on action-perception correlation, we document mechanisms underlying the emergence of these assemblies, especially (i) the recruitment of neurons and consolidation of connections defining the kernel of the assembly along with (ii) the pruning of the cell assembly's halo (consisting of very weakly connected cells). We found that, whereas a learning rule mapping covariance led to significant overlap and merging of assemblies, a neurobiologically grounded synaptic plasticity rule with fixed LTP/LTD thresholds produced minimal overlap and prevented merging, exhibiting competitive learning behaviour. Our results are discussed in light of current theories of language and memory. As simulations with neurobiologically realistic neural networks demonstrate here spontaneous emergence of lexical representations that are both cortically dispersed and anatomically distinct, both localist and distributed cognitive accounts receive partial support.

What can autism teach us about the role of sensorimotor systems in higher cognition? New clues from studies on language, action semantics, and abstract emotional concept processing

“…Note that the production of syllables, as it dominates the stage of repetitive babbling, implies the activation not only of neurons in frontal articulatory motor areas (where speech output is controlled) but at the same time of auditory neurons in posterior temporal areas responding to the self-produced sounds; similar co-occurrence of activity is present in somatosensory fields in anterior parietal cortex. Such babbling-related co-occurrence of neuronal activity has been shown (by computer model simulations [Garagnani et al, 2008;Garagnani, Wennekers, & Pulvermü ller, 2009]) to yield circuits that interlink motor and sensory neurons. Because they interlink information about actions and their related perceptions, we call these circuits 'action perception circuits'.…”

Neural plasticity and concepts ontogeny

“…3.3 is applied in simulating visual areas V1, V2, VO and LOC, and the auditory areas A1 and STS, demonstrating various aspects of lexical categorization, like fast mapping (Swingley 2010). Garagnani et al (2000) used a similar basic model of cortical maps with lateral excitatory and inhibitory connections, in reproducing with a finer degree the auditory pathway, including A1, auditory belt and parabelt areas, inferior prefrontal, premotor and primary motor areas, in an experiment of word learning based on action-perception correlation. Fuster (2001Fuster ( , 2008 has influentially emphasized that the entire cortex is organized in hierarchical manner, with levels of representation that are progressively more integrative and abstract.…”

Learning visual-motor Cell Assemblies for the iCub robot using a neuroanatomically grounded neural network

“…1.B. In this section we present only a brief overview of salient features of the model-the mathematical formulation of the model, structural and connectivity details, and learning mechanisms, have been previously described elsewhere [9,[13][14][15][16] . We implemented a network consisting of six areas of artificial neurons (graded response cells) with reciprocal connections between and within areas.…”

“…In previous work [9,[13][14][15][16] we have shown that networks that include this range of realistic neurobiological features tend to exhibit the formation of (input-specific) memory circuits, corresponding to what Hebb once postulated and labelled "Cell Assemblies" [17] or CAs. Below we identify the brain areas that the model simulated, and review the neuroanatomical evidence that we used as a basis to establish the presence, in the network, of direct links between pairs of such areas.…”