Somatodendritic consistency check for temporal feature segmentation

Toshitake Asabuki,Tomoki Fukai

doi:10.1038/s41467-020-15367-w

Abstract

The brain identifies potentially salient features within continuous information streams to process hierarchical temporal events. This requires the compression of information streams, for which effective computational principles are yet to be explored. Backpropagating action potentials can induce synaptic plasticity in the dendrites of cortical pyramidal neurons. By analogy with this effect, we model a self-supervising process that increases the similarity between dendritic and somatic activities where the somatic activity is normalized by a running average. We further show that a family of networks composed of the two-compartment neurons performs a surprisingly wide variety of complex unsupervised learning tasks, including chunking of temporal sequences and the source separation of mixed correlated signals. Common methods applicable to these temporal feature analyses were previously unknown. Our results suggest the powerful ability of neural networks with dendrites to analyze temporal features. This simple neuron model may also be potentially useful in neural engineering applications.

Highlights

The brain identifies potentially salient features within continuous information streams to process hierarchical temporal events
A family of competitive networks of the proposed neuron model can perform a variety of unsupervised learning tasks ranging from chunking to blind source separation (BSS), which were previously performed by specialized, distinct networks and learning rules
Our model learns temporal features of an input based on a novel learning rule which we call minimization of regularized information loss (MRIL)

Summary

Introduction

The brain identifies potentially salient features within continuous information streams to process hierarchical temporal events. In predictive coding[7,8,9], the brain may chunk information in bottom-up and top-down pathways to identify variables relevant to the hierarchical Bayesian modeling of mental processes Another important class of temporal feature analysis is blind source separation (BSS: related to the so-called cocktail party effect) in which the brain separates mixed sensory signals (typically auditory) from multiple sources in order to recognize the individual sources[10]. In a twocompartment neuron model, that the minimization of information loss between dendritic synaptic input and a neuron’s own output spike trains enables efficient learning of clustered temporal events in a completely unsupervised manner. Our formulation is inspired by the observations that neuronal adaptation shifts the neuron always toward a regime of efficient information transmission[16,17,18,19]

Methods

Results

Conclusion