The Receptive-Field LISSOM model demonstrates that the only major requirements for cortical self-organization are short-range excitation, long-range inhibition, and Hebbian weight adaptation. Starting from random-strength connections, neurons develop localized receptive fields and lateral interaction profiles cooperatively and simultaneously. Self-organization takes place with multiple retinal inputs, and with multiple anatomical receptive fields of varying sizes, but requires either large initial receptive fields or ordered receptive field centers.
The model explains why lateral connection patterns in the visual cortex follow the patterns of receptive field properties. As receptive fields organize in the cortex, specific patterns of activity correlations appear in the cortical map. Lateral connections store these correlation patterns. Such correlations also form the basis for Gestalt knowledge such as continuity of contours, proximity, and smoothness [17,13,12]. By analyzing the patterns of lateral connections, it may be possible to determine what Gestalt information is actually represented in the cortex. In future research, we intend to study how Gestalt knowledge in the lateral connections may be used for feature grouping and segmentation.