Computational neuropsychology 2.0: A deep learning account of the pattern of deficits in visual recognition after brain damage

Localized brain damage can result in relatively specific problems inmental faculties, including problems with visual recognition. It is still amatter of debate why certain deficits and combinations of deficitsoccur. Recent evolutions in human neuroscience, including our ownwork, have revealed the complexity of the system that supportsvisual recognition, and models such as deep neural networks (DNNs)have offered us a computational understanding of many properties ofthis system. This empirical knowledge and computational insightshave high potential for a better understanding of neuropsychologicaldeficits, and, in the other direction, neuropsychological cases offer aunique approach to validate the available computational models. Inthis project we will apply this new approach, here referred to ascomputational neuropsychology 2.0. We will develop and validate anew test battery of recognition tasks including multiple categoriesand a manipulation of recognition difficulty. We will show howperformance associations between object categories in healthyindividuals and neural overlap in the healthy brain is related torepresentational overlap in DNNs. Crucially, we will test a largesample of stroke patients, and show to what extent the pattern ofdeficits across object categories and across difficulty levels can beexplained by the information processing in DNNs, allowing us to'mimic' the full spectrum of human recognition deficits in artificialintelligence models.