Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.799999
Title: Uncovering temporal structure in neural data with statistical machine learning models
Author: Higgins, Cameron
ISNI:       0000 0004 8507 1368
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Methods for analysing neural dynamics have predominantly focussed on resting state activity, typically using unsupervised models to infer the main modes of spontaneous variation in the absence of deliberately controlled experimental stimuli. Analysis of neural task data, on the other hand, overwhelmingly assumes responses that are perfectly aligned in time and synchronised over multiple trials by the onset time of the stimulus. These assumptions contradict widespread evidence for a more dynamic, time-varying structure of neural responses. We here explore methods to relax these assumptions and explicitly model the temporal variability of neural task data, by exploring both supervised and unsupervised extensions of the Hidden Markov Model framework. We establish the importance of this approach by applying an existing model - previously introduced for resting state analysis - to a task-based experiment, demonstrating the existence of distributed patterns of variation over trials that reflect underlying fluctuations of cognitive state and predict behavioural responses. We then develop novel supervised frameworks for analysing faster dynamics directly associated with stimulus processing, and show through simulations how these models improve performance where assumptions of temporal synchrony and alignment over trials are not met. Applying this to a MEG experiment on replay, we prove that these models better extract the successive stages of visual stimulus processing in the data, and then that this improves sensitivity to detecting replay across multiple stages of the visual hierarchy. This allows us to investigate the temporal patterns of activation across the visual hierarchy, and we show that spontaneous replay reverses the original feed-forward direction of information flow, with replayed representations in higher visual areas reactivating ahead of the associated lower level visual areas. Finally, we return to the original unsupervised model to compare how the onset of replay events aligns to whole-brain patterns of activity, and show that replay occurs selectively during periods of default mode network activation. These significant findings advance our understanding of the neural physiology underlying replay, whilst also demonstrating the potential of the novel methodology that we have developed.
Supervisor: Woolrich, Mark ; Behrens, Timothy Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.799999  DOI: Not available
Keywords: Neurosciences ; Bayesian Modelling ; Machine Learning
Share: