Causal feature transfer in different face categorisation tasks (360G-Wellcome-202585_Z_16_Z)

The research seeks to study how the causal feature transfer of face representations in the brain may depend on the categorisation task itself. Subjects would perform two face discrimination tasks – one for gender and one for the six basic emotional expressions (anger, disgust, fear, happiness, sadness, surprise) – with stimulus pixels sampled using the Bubbles method. MEG measures would be used to examine in a first stage which nodes are maximally task-modulated by computing the time-varying Mutual Information between MEG source activity and tasks. For each of these task-sensitive nodes, we would then compute the Mutual Information between the sampled pixels and the MEG node activity. Between these two analyses we could separate, in the set of task-varying nodes, the subset of nodes coding categorization-specific features, from those coding only the semantics of the categorization task. In a final stage, we would reconstruct the network of nodes that code and transfer task-related features. From a theoretical perspective, this latter step is important to dissociate a code that the experimenter can decode, from a code that the brain uses for data transmission.