Fig.1
Shigeru Tanaka
Laboratory for Visual Neurocomputing, RIKEN Brain Science InstituteIt is accepted that visual perception is achieved by grouping and segmentation of a particular visual object from complex patterns presented in the retina (Fig. 1a). Recent findings from electro-physiological studies suggest that synchronous spike activities among different visual cortical neurons are involved in the grouping and segmentation processes for visual perception. The functional role of correlated activities of cortical neurons draw special attention. So far we have successfully demonstrated qualitatively a mechanism of tilt illusion (e.g., Fig. b), using a large-scale network composed of simple neuron units in the rate-coding scheme. However, the neuron units in the rate-coding scheme are no longer satisfactory to understand a relationship between spike-based correlated activities and visual perception. We need to build a network model composed of more realistic neuron units that elicit spike activities in response to spatiotemporal visual patterns presented in the model retina. In this project, we attempt to construct such a network model of the visual cortex by integrating many lines of evidence obtained from psychophysics, neurophysiology, anatomy etc. towards the understanding of a holistic mechanism of visual perception.
Now we are building a large scale neural network model composed of integrate-and fire units taking into account intracortical connections as well as afferent inputs from the LGN (Fig. 2) to seek for neural correlates of feature binding and segmentation. In this model, geniculo-cortical inputs are obtained by our self-organization model, and intracortical connections are assumed to be short-range excitatory and long-range-inhibitory isotropic connections. From the computer simulation based on this model, we found that orientation and direction selectivity is enhanced by the intracortical connections though the selectivity emerges from the pattern of afferent inputs (Fig. 3).
