B1078
Title: Exploring dynamic factors of fMRI activity in the presence of sparse loadings
Authors: Tak-Shing Chan - Lancaster University (United Kingdom)
Xinle Tian - University of Bath (United Kingdom)
Kai Zheng - Lancaster University (United Kingdom)
Alex Gibberd - Lancaster University (United Kingdom) [presenting]
Abstract: Recent research is presented looking into the application of dynamic factor models to fMRI data. By considering an expectation-maximisation approach to estimation, a regularised likelihood is introduced that encourages sparsity in the factor loadings. It enables enhanced interpretation of the resultant factors, and in the context of fMRI provides spatial localisation of the factors. The results of this approach are contrasted, where latent dynamics, and dependence, are explicitly modelled to canonical ICA approaches which assume independence. The framework is further extended to the group-sparse setting and whether anatomical priors are useful for describing, in a predictive sense, fMRI activity is considered.
