UC Santa Barbara

The hippocampal formation plays an important role in spatial navigation and episodic memory. At the core of the neuroscience community's understanding of this role, are classes of functional cell types, such as place cells and grid cells. Despite being the subject of intense study for over half a century, how the complex recurrent circuitry of the hippocampal formation generates these functional classes, and how they interact with each other to support the critical cognitive processes the hippocampal formation underlies, is an area of open research. In this thesis, we present experimental and computational work aimed at addressing some aspects of these questions.

First, we develop a novel technique to optically access the transverse plane of the hippocampus in vivo, allowing - for the first time - the neural activity of multiple subregions of the hippocampus to be simultaneously recorded. We utilize this tool to characterize functional and structural properties across the hippocampal circuit, finding stable and unstable populations of dendritic spines on the apical dendrites of CA1 pyramidal neurons and heterogeneity in the amount of spatial information encoded by place cells along the CA1-CA3 axis. Second, we test a long held assumption on properties of grid cells in medial entorhinal cortex, analyzing recent cutting-edge experimental data. We find evidence for the distribution of grid spacings and orientations in individual modules to be non-uniform, having small, but significant variation. Computational modeling leads us to conjecture that a grid code with such heterogeneity in its properties enables robust encoding of local spatial information. And third, we study the emergence of localized responses in artificial neural network models. We find that, as these local features form, the statistics of the internal representations of the network become increasingly non-Gaussian, in-line with theoretical work that has suggested this to be a general mechanism for driving localized responses. Sparsifying the network, via pruning or regularization, amplifies the non-Gaussian statistics, emphasizing the role of sparsification on internal representations. Each of these experimental and computational approaches motivates exciting avenues of future direction to shed light on the computations performed by the hippocampal formation.