Complex sound representation in the brain: receptive field properties of neurons and astrocytes in the auditory cortex
File(s)
Author(s)
Lu, Sihao
Type
Thesis
Abstract
In order to understand how the brain achieves both selectivity and invariance, there is a need to investigate the features present in natural stimuli and how cells respond to these features. This work aims to characterise the receptive field properties of both neurons and astrocytes within the auditory cortex in mice. Whilst extensive work has been conducted in the characterisation of neuronal receptive fields using parametric stimuli such as pure tones, there exists far less work probing neuronal circuits with natural vocalisations. I present the receptive field properties of single neurons in the mouse auditory cortex in response to both ethologically relevant stimuli in the form of ultrasonic vocalisations, and irrelevant stimuli, in the form of pitch-shifted bird song.
In addition to neurons, there is a growing body of evidence pointing to the involvement of astrocytes in sensory processing. Astrocytes tile the brain and are each tightly connected with thousands of neurons. In this work I explore how these cells can encode sensory stimuli in the form of Ca2+ fluctuations using genetically encoded calcium indicators and two-photon microscopy. I show that astrocyte calcium activity is modulated by multiple features of a naturalistic stimulus. Moreover, I show that this modulation can occur on the scale of sub-cellular compartments in the microdomain. Lastly, I examine how the inclusion of biological properties into artificial object recognition systems can confer improved performance. I show that the inclusion of flexible neuron-to-computation into modern convolutional neural network architectures results in improvements against adversarial examples and improves performance when dealing with low amounts of data.
In addition to neurons, there is a growing body of evidence pointing to the involvement of astrocytes in sensory processing. Astrocytes tile the brain and are each tightly connected with thousands of neurons. In this work I explore how these cells can encode sensory stimuli in the form of Ca2+ fluctuations using genetically encoded calcium indicators and two-photon microscopy. I show that astrocyte calcium activity is modulated by multiple features of a naturalistic stimulus. Moreover, I show that this modulation can occur on the scale of sub-cellular compartments in the microdomain. Lastly, I examine how the inclusion of biological properties into artificial object recognition systems can confer improved performance. I show that the inclusion of flexible neuron-to-computation into modern convolutional neural network architectures results in improvements against adversarial examples and improves performance when dealing with low amounts of data.
Version
Open Access
Date Issued
2022-04
Date Awarded
2022-09
Copyright Statement
Creative Commons Attribution NonCommercial NoDerivatives Licence
Advisor
Kozlov, Andriy
Schultz, Simon
Clopath, Claudia
Sponsor
Engineering and Physical Sciences Research Council
Publisher Department
Bioengineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)