Dr Marieke Mur, MRC Cognition & Brain Sciences Unit, University of Cambridge

Modelling high-level object representations in brain and behaviour

Object representations in human high-level visual cortex are at the interface between perception and cognition. What is the nature of these representations, and how are they computed? Furthermore, can they predict human behaviour? I will show that object representations in high-level visual cortex are at once categorical and continuous, and can be explained similarly well by category labels, visual features of intermediate complexity, and deep convolutional neural networks. Among the visual features, it is those correlated with category membership that explain the high-level visual object representation. I will further show that the high-level object representation predicts human object-similarity judgments reasonably well, but fails to capture evolutionary more recent category divisions present in the judgments. Together, these findings suggest that high-level visual cortex has developed feature detectors that distinguish between categories of long-standing evolutionary relevance, and that other brain systems might adaptively read out or introduce category divisions that serve current behavioural goals.