Random neuronal ensembles can inherently do context dependent coarse conjunctive encoding of input stimulus without any specific training

Jude Baby George,Sujit Kumar Sikdar,Zubin Rashid,Bharadwaj Amrutur,Grace Mathew Abraham

doi:10.1038/s41598-018-19462-3

Jude Baby George, Sujit Kumar Sikdar + Show 3 more

Open Access

https://doi.org/10.1038/s41598-018-19462-3

Copy DOI

Abstract

Conjunctive encoding of inputs has been hypothesized to be a key feature in the computational capabilities of the brain. This has been inferred based on behavioral studies and electrophysiological recording from animals. In this report, we show that random neuronal ensembles grown on multi-electrode array perform a coarse-conjunctive encoding for a sequence of inputs with the first input setting the context. Such an encoding scheme creates similar yet unique population codes at the output of the ensemble, for related input sequences, which can then be decoded via a simple perceptron and hence a single STDP neuron layer. The random neuronal ensembles allow for pattern generalization and novel sequence classification without needing any specific learning or training of the ensemble. Such a representation of the inputs as population codes of neuronal ensemble outputs, has inherent redundancy and is suitable for further decoding via even probabilistic/random connections to subsequent neuronal layers. We reproduce this behavior in a mathematical model to show that a random neuronal network with a mix of excitatory and inhibitory neurons and sufficient connectivity creates similar coarse-conjunctive encoding of input sequences.

Highlights

Pattern or sequence recognition and classification is a well-studied problem in engineering that uses biologically inspired architectures like artificial neural networks, and more recently deep learning networks that have shown promising results in solving such tasks
We show that responses from a neuronal ensemble grown on multi-electrode array show coarse-conjunctive encoding of multiple spatio-temporal inputs and demonstrate their ability to do context dependent encoding, which can be decoded/classified robustly using ‘perceptrons’ as proxy for the output neuron shown in layer 3 (L3) of Fig. 1
All animal experiments were performed in accordance with guidelines, rules and regulations of the Institutional Animal Ethics Committee (IAEC) for animal experiments of the Indian Institute of Science, Bangalore, India constituted as per article number 13 of the CPCSEA (Committee for the purpose of Control and Supervision of Experiments on Animals, http://cpcsea.nic.in) rules, laid down by Government of India

Summary

Introduction

Pattern or sequence recognition and classification is a well-studied problem in engineering that uses biologically inspired architectures like artificial neural networks, and more recently deep learning networks that have shown promising results in solving such tasks. We describe this further in the schematic, where a layered neuronal system with probabilistic connectivity at input and output of first layer, is connected to a second layer having neurons equipped with STDP, to solve the problem of input classification without any need for network modification/learning at the input layer We experimentally validate this architecture by using neuronal ensembles cultured on a multi electrode array, to form the first layer of the Fig. 1. This is illustrated by single neurons coding for 3 different shapes (square, triangle and circle) and 3 different colors in the first layer (L1). The way information is encoded and decoded across different layers before it converges on the output neuron is not known

Methods

Results

Conclusion