A two-stage approach to estimating voxel-specific encoding models
No Thumbnail Available
The process of reconstructing the activity of a brain in response to a certain stimulus is no trivial task. Building on the work by G u cl u & van Gerven, 2014  this thesis aims to improve the presented model by using (i) high-resolution stimuli in order not to lose information by downsizing the stimuli, (ii) using only the receptive elds of the single voxel (volumetric pixel) responses since the redundant areas could cause over- tting, (iii) using convolution to minimize artefacts and (iv) relate the sizes of the features to actual neuron receptive eld sizes. We show that an approach that learns the receptive elds from the data itself improves the prediction accuracy signi cantly.
Faculteit der Sociale Wetenschappen