Learning inhibition-controlled sequences through unsupervised plasticity in spiking neural networks

Keywords

Loading...
Thumbnail Image

Issue Date

2022-11-28

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

The brain can learn the temporal order of events through exposure to sequential information. This can lead to the emergence of feedforward structures within cortical circuits. Understanding the neurobiological mechanisms that support the generation of such structures can deepen the knowledge of the learning and memory processes. Previous computational approaches have addressed sequence learning in spiking neural networks through plasticity between excitatory neurons. However, to do so, they rely on handcrafted connectivity or nonbiological assumptions. Recent studies have indicated that circuits of inhibitory interneurons might control several functions by regulating the activity of excitatory cells. Here, I propose a model for learning inhibition-controlled sequences through the interaction of excitatory and inhibitory spike-timing dependent plasticity (STDP). The proposed network consists of one excitatory, E, and two inhibitory populations, I1 and I2, that are responsible for the formation of excitatory assemblies and their arrangement in sequential order, respectively. In contrast to other models of sequence learning, here, learning is realized through unsupervised plasticity principles alone. I show that plasticity on the I1-to-E, I2-to-E, and E-to-I2 synapses are essential for inhibition-controlled sequence learning. Adjusting the timescales and learning rate of inhibitory STDP and regulating the activity of I2 neurons contribute to the reliable replay of sequences. Plasticity between E and I2 neurons during spontaneous background activity after learning degrades the learned connectivity and can lead to sequence memory loss. The model shows that interneuron circuits can have a functional role in memory by learning to control the propagation of activity between excitatory cells. Moreover, it suggests that including multiple populations of interneurons with distinct functions increases neurobiological realism compared to previous models of sequence learning.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen