The project’s researchers – at
Results from psychophysics, neuroscience and computational modelling show that the rapid recognition of everyday objects can be explained by viewing the brain’s visual cortex as a multi-layer, feed-forward system in which the neural activity propagates from the eye to the higher brain areas, with little feedback from the higher layers to the lower layers.
Yet, there are as many feedback connections as feed-forward connections in the visual cortex, and the researchers will seek to understand their role.
‘This project will enhance our understanding of this process by drawing on the recent progress in a new class of machine learning methods called ‘deep belief networks,’ and through new experimental methods to study the visual cortex,’ said Yann LeCun, a professor of computer science at the Courant Institute.
‘Learning algorithms for deep belief networks could help us to model how the visual cortex learns because they can be applied to multi-layer architectures similar to the visual cortex, and because feedback connections play a crucial role in the learning process in these models,’ added LeCun.