Dr Ian Charest | Tracking object representations in the brain

Information about the visual world is represented in the brain in an individually unique way. Honouring these representational idiosyncrasies is paramount if we are to gain a thorough understanding of object recognition. In this talk, we will follow these representations in the visual ventral stream, from early low-level feature extraction all the way until they are readily available in consciousness. In the first part of the talk, I will show how we use representational similarity analysis (RSA) to track idiosyncratic object representations in space (fMRI) and time (EEG). RSA consists of constructing representational dissimilarity matrices (RDMs) from pattern distances. These RDMs can characterise behavioural, cortical or computational representational spaces. Importantly, these RDMs can be combined across measurement modalities, opening the door to unprecedented explanatory power in cognitive neuroscience. Using similarity-based fusion of fMRI and EEG, we investigate how personally meaningful and unfamiliar objects follow different spatio-temporal trajectories in the brain. In the second part of the talk, I will present new results looking into conscious access to visual information. To study conscious access, we use deep convolutional neuronal networks (DCNN) and brain imaging (fMRI, EEG) to model conscious access measured behaviourally with an attentional blink task. We find that visual categories differ in their access to consciousness, with animate objects presenting markedly reduced attentional blink magnitude. DCNNs can be used to model visual features of increasing complexity, and these visual features can explain a substantial amount of the variance between objects in their conscious access. In addition, we show that target-target similarity interacts with target selection, providing a mechanistic explanation of the AB phenomenon and of conscious access in object recognition.