An Indexing Theory for Working Memory Based on Fast Hebbian Plasticity

Abstract

Working memory (WM) is a key component of human memory and cognition. Computational models have been used to study the underlying neural mechanisms, but neglected the important role of short-term memory (STM) and long-term memory (LTM) interactions for WM. Here, we investigate these using a novel multiarea spiking neural network model of prefrontal cortex (PFC) and two parietotemporal cortical areas based on macaque data. We propose a WM indexing theory that explains how PFC could associate, maintain, and update multimodal LTM representations. Our simulations demonstrate how simultaneous, brief multimodal memory cues could build a temporary joint memory representation as an "index" in PFC by means of fast Hebbian synaptic plasticity. This index can then reactivate spontaneously and thereby also the associated LTM representations. Cueing one LTM item rapidly pattern completes the associated uncued item via PFC. The PFC-STM network updates flexibly as new stimuli arrive, thereby gradually overwriting older representations.

Traces of semantization - from episodic to semantic memory in a spiking cortical network model

“…The network model used here features two reciprocally connected networks, the so-called Item and Context networks. The architecture of each network follows our previous spiking implementations of attractor memory networks (Lansner, 2009; Tully et al, 2014, 2016; Lundqvist et al, 2011; Fiebig and Lansner, 2017; Chrysanthidis et al, 2019; Fiebig et al, 2020), and is best understood as a subsampled cortical layer 2/3 patch with nested hypercolumns (HCs) and minicolumns (MCs; Fig. 1A, see STAR⋆METHODS for details).…”

“…Recently we demonstrated that the same model, enhanced with a Bayesian-Hebbian learning rule (Bayesian Confidence Propagation Neural Network, BCPNN) to model synaptic and intrinsic plasticity, was able to quantitatively reproduce key behavioral observations from human word-list learning experiments (Fiebig and Lansner, 2017), such as serial order effects during immediate recall. This model performed one-shot memory encoding and was further expanded into a two-area cortical model used to explore a novel indexing theory of working memory, based on fast Hebbian synaptic plasticity (Fiebig et al, 2020). In this context, it was suggested that the underlying naive Bayes view of association would make the associative binding between two items weaker if one of them is later associated with additional items.…”

Embodied working memory during ongoing input streams

“…Short-term synaptic plasticity has been supported as a candidate mechanism for storing information in WM [ 24 , 25 ]. Moreover, fast-expressing Hebbian synaptic plasticity may modify network connectivity momentarily enough to support information storage in WM [ 74 – 76 ]. Arguably, Hebbian forms of synaptic plasticity are incompatible with the flexible functionality of WM, because they induce long-lasting changes in synaptic connections that generally outlast the duration of persistent activity.…”

Mapping the BCPNN Learning Rule to a Memristor Model

“…Importantly, both reduced non-spiking and biologically detailed spiking realizations of BCPNN perform similar functions. They have been extensively used to model brain-like cognitive capabilities such as associative memory (Johansson and Lansner, 2007 ; Lundqvist et al, 2011 ), episodic memory (Chrysanthidis et al, 2021 ), and working memory (Fiebig and Lansner, 2017 ; Fiebig et al, 2020 ), which play a key role in human intelligence. In a broader perspective, we suggest that these advancements in simulating different aspects of human cognitive function within a system framework of brain-like BCPNN constitute a promising direction in the development of artificial general intelligence (AGI).…”