Testing theories of computation and learning in the visual cortex

Thesis

Modern systems neuroscience provides experimental access to the dynamics of neural systems over different timescales. Many studies of sensory systems in the brain do not exploit this and have instead focused on on static features measured from stimulus-evoked responses. In this thesis, we show that observation of dynamics – changes in neural variables over time – can offer a means to distinguish theoretical models relating to key unsolved questions in neuroscience. We apply this approach to two types of dynamical systems model. In the first study, we analyse the dynamics of firing rates in model recurrent networks. We offer routes to experimentally testing particular implementations of theories of sensory processing such as efficient and predictive coding. In the second study, we propose that learning dynamics – the evolving patterns of neural changes across time compared within and across areas – can serve as a crucial hallmark to test theories of learning in the brain. We devise a multi-stage behaviour and neuronal imaging paradigm in mice to experimentally test key predictions of the deep learning theory of perceptual learning and ultimately find that several key predictions were not met. Taken together, the work in this thesis serves as a demonstration of how experiment and theory in neuroscience can be combined with careful consideration. Through testing a popular theory, it also highlights key challenges in this approach.

Authors: Armstrong, SL

• DOI: 10.5287/ora-mpmobvqvv
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: visual cortex; perceptual learning; orientation tuning; deep learning; systems neuroscience; learning dynamics; vision; 2-photon imaging; predictive coding
• Subjects: Multiphoton excitation microscopy; Perceptual learning; Computational neuroscience; Neural circuitry; Calcium imaging