Learning optimizes decision templates.

Classification image analysis has been used successfully for extracting the critical image features that observers use when making perceptual judgments. Using this approach, previous studies have shown that learning re-tunes decision templates and shapes task-relevant features. Although classification images have been used widely in psychophysics, the application of this technique to fMRI studies has been limited due to noise in single-trial fMRI responses. In this study, we aim to develop a novel methodology that uses multi-voxel pattern analysis to extract classi?cation images from fMRI data and reveal the neural representation of behaviorally relevant features. We investigate how learning re-tunes the representation of visual features in the human visual cortex according to their behavioral relevance for perceptual decisions. This work will provide 1) an effective way to extract classification images from fMRI data; 2) significant insights into the role of learning in tuning feature representations in the human brain.