Use this URL to cite or link to this record in EThOS:
Title: Modelling environmental and self-motion influences on grid cell firing
Author: Evans, Talfan
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Grid cells have been suggested to play a role in path integration, the process of integrating self-motion cues in order to maintain an estimate of location relative to a previous estimate. However, a system performing pure path integration will integrate error such that the estimate of location will drift over time. In contrast, grid cells are known to be stable over time suggesting that they are stabilized by sensory inputs from their environment. Together with their striking anatomical arrangement, this observation also suggests that grid cells may constitute an efficient encoding of space that might be useful for localization or navigation over large distances. However, distortions to the canonical hexagonal pattern have been observed in large (Stensola et al. (2015)) or non-rectangular (Derdikman et al. (2009); Krupic et al. (2015)) environments. The same ring patterns are also known to be non-stationary over time, temporarily rescaling in novel environments (Barry et al. (2012)), rescaling in response to environmental manipulations (Barry et al. (2007); Stensola et al. (2012); Krupic et al. (2018)), undergoing a shear-like transformation and re-alignment with experience of a square box (Stensola et al. (2015); Krupic et al. (2015)) and adjusting local representations to reflect global structure (Carpenter et al. (2015)). Sensory input to grid cells is thought to be mediated by place cells or boundary vector cells found in areas CA1/3 of the hippocampus and subiculm, respectively. Existing models rely on strong stimuli from pre-learned associations with these sensory inputs to `reset' the accumulated error in the grid pattern (Fuhs and Touretzky (2006); Burak and Fiete (2009); Pastoll et al. (2013); Hardcastle et al. (2015)). However, this `hard-resetting' mechanism can lead to localization errors when sensory and path integration estimates mismatch and does not support efficient learning of these sensory associations in novel environments. In the first part of this investigation, I ask whether principles of optimality based on the geometric properties of the grid cell ring pattern can explain the observed orientation offset and accompanying shearing with experience. Later, I develop these ideas in the context of continuous attractor models, a class of mechanistic models based shown to reproduce several characteristics of grid cells. In the second part of this investigation I propose a novel, biologically plausible model of sensory integration in grid cells. The resulting system is capable of simultaneously navigating and learning sensory associations in novel environments and is shown to account for several unexplained experimental observations. Next, I augment this system with the ability to perform o ine inference of learned spatial associations. I propose that this inference is achieved by a process of message passing on a graph of learned locations (represented by place cells) and propose a novel biological mechanism. By allowing the system to schedule updates to learned locations based on the local convergence of the graph, I show that the process of offline inference generates sequences of reactivations which bear resemblance to the phenomenon of hippocampal replay. Lastly, I conduct a novel analysis of local distortions to the grid pattern, relating the findings to theoretical work presented earlier in the thesis.
Supervisor: Burgess, N. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available