Through massively parallel computational simulations, we studied how a large network of simple neural elements (the RF-LISSOM model) could develop a functional organization similar to that of the primary visual cortex. It was found that starting from a 'tabula rasa' state, the afferent and lateral connections in the network self-organized cooperatively and simultaneously through a common Hebbian mechanism, and produced receptive fields (RFs), orientation maps, and patterns of lateral connections that follow the receptive field organization. Second, we hypothesized that similar self-organizing mechanisms continue operating in the adult cortex, maintaining it in a continuously-adapting dynamic equilibrium with the input, and tested this hypothesis on the self-organized model. When the equilibrium was perturbed by a retinal scotoma, RFs expanded in size in a reversible fashion, matching recent neurobiological observations in the cat and psychophysical experiments in the human. Third, a possible functional role for the lateral connections in the cortex was verified in the model. The lateral connections learned correlations in the network activity, and in processing retinal input, filtered out redundancies and established a sparse coding of the input. The conclusion is that the lateral connections in the cortex could act as as a negative filter that allows the cortex to efficiently process the massive amounts of visual information presented by the environment.
|Title of host publication||Proceedings of the International Conference on Artificial Neural Networks|
|Editors||Joseph Sirosh, Risto Miikkulainen, Yoonsuck Choe|
|Place of Publication||Berlin|
|Number of pages||6|
|Publication status||Published - 1996|