Jafarpour, A; (2014) Decoding oscillatory representations of visual stimuli in episodic memory and working memory. Doctoral thesis , UCL (University College London).

Preview

PDF corrected_thesis_Anna Jafarpour.pdf Available under License : See the attached licence file. Download (18MB)

Abstract

Theories inspired by electrophysiological studies in animals suggest that the replay of past experiences plays an important role in episodic memory as well as working memory; yet very little is known about the neural characteristics of such replay in the human brain. This thesis consists of neuroimaging experiments for studying the temporal characteristics of the replay in the human brain and analytical methods for decoding replay. To that end, oscillatory neural activity patterns were recorded from healthy young adults via a non-invasive electrophysiological technique (Magnetoencephalography, MEG). Firstly, a pipeline for decoding MEG data using machine learning algorithms was eveloped and proposed. Then using an associative recognition experiment, we marked the neural signature for categorical visual information (about faces and scenes) during encoding. These markers of encoding-related experiences were then used for detecting the replay during retrieval - triggered by an associative memory cue. As a result, replay was detected at about 500 ms from onset of the cue. The results suggest that episodic recollection and replay are accomplished within 500 ms. Next, in a working memory experiment, I used item speci c visual information for tracking the replay of oscillatory activity while maintaining that information. Three visual stimuli with presumably distinct cortical representations were selected (types: a face, a banana, and a chair) and presented in a sequential order. Event-related responses during encoding showed a main e ect of item type and working memory load at 400-500 ms from onset of the stimuli. Using a decoding approach, it was possible to categorize oscillatory patterns related to each of the three stimulus types. These decoders are now used as markers of item speci c replay in working memory during the maintenance phase. This analysis is ongoing. Finally, we proposed a pipeline for detecting an optimal feature space for decoding MEG data at a group level because the previous pipeline relied on di erent features across subjects for decoding. Here the Canonical Variates Analysis of beamformer reconstructed MEG data in source space was used. Canonical Variates Analysis stimated the dependency of the selected features of MEG data to the experimental conditions and enabled multivariate decoding of MEG signal in the source space. Thus this proposed method was an optimal way for group level inference of MEG multivariate analysis. Overall, the MEG based decoding of the representation of visual stimuli was shown in source and sensor spaces. Also, our results revealed the temporal characteristic of replay in an episodic memory experiment.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item