Cognitive systems in the brain: modelling invariant object recognition in natural visual environments. (360G-Wellcome-079174_Z_06_Z)

We will build on recent neurophysiological discoveries on inferior temporal visual cortex neuronal activity to develop understanding at the computational neuroscience level of the mechanisms that underlie our ability to recognise objects in natural scenes, independently of their view, size, and exact position. Novel goals include: 1. We will incorporate our discoveries on the receptive fields of neurons in natural scenes into a refined model to show howinvariances can be learned, and how multiple objects can be recognised, in natural scenes. 2. We will incorporate a new 'continuous transformation' mechanism for invariance learning that we have recently discovered. 3. Becausethe spatio-temporal statistics of the world interact with the functional architecture and dynamics of the visual system, we will use 3D modelling toolsto generate training images with spatio-temporal statistics that are quantitatively defined but also increasingly approach those present when natural scenes are viewed. 4. We will correspondingly add realistic dynamics to the system by introducing integrate-and-fire neurons to some simulations.