Computer Vision News - July 2020

2 Summary Research 4 This is the first time that a deep network has predicted a feature of the brain - unknown before - and turned out to be true. This impressive result may help soon the scientists figure out how the primate brain solves the object recognition problem. Reducing the brain - or in the case of Plato’s philosophy, human nature - in its fundamental characteristics is an extremely difficult problem, troubling philosophers since the antiquity. This reduction problem is the basis for pattern identification in data based on the correlation between their features, such as in principal component analysis (PCA). From this research, deep learning and neuroscience are approached in a dualistic perspective to prove that the brain uses a mathematical categorization of a visual object according to their principal components. Those are stored in a two- dimensional map of cells which represent those different objects. Patterns are stored in cell locations inside the map which further determine the features of preferred objects. Similarly, to a convolutional neural network, common shapes are categorized together, such as round objects like faces or oranges. The brain's inferotemporal (IT) cortex is crucial for visual object recognition and it is considered the ultimate stage of the vertical cortical visual system. Different Every month, Computer Vision News reviews a research paper from the AI field. This month we have a very exciting article, which was published less than a few days ago in Nature: A map of object space in primate inferotemporal cortex. The paper explores how the brain represents visual objects, utilizing deep learning architectures. We are indebted to the authors (Pinglei Bao and Doris Y. Tsao are the main authors, with co-authors the postdoctoral A map of object space in primate inferotemporal cortex by Ioannis Valasakis @wizofe scholar Liang She and the graduate student Mason McGill), for allowing us to use their images to illustrate this review. You can find their paper at this link.