A neural network model for the simultaneous self-organization of topographic receptive fields and lateral interactions in cortical maps is presented. The afferent and lateral connection weights in the network are initially random, but self-organize based on external input to form topographic maps. The maps are in dynamic equilibrium with the input, and can reorganize in response to lesions in the network. During reorganization, the area of functional loss resulting from the lesion first increases as lateral connections adapt, and then decreases as afferent connections reorganize to compensate. The reorganizing behavior closely matches experimental observations on cortical lesions and stroke. The model shows how lateral interactions produce dynamic receptive fields and predicts that adapting lateral interactions are fundamental to cortical reorganization. Based on the model, two techniques to accelerate recovery from stroke and cortical surgery are suggested.

PDF

In Computational Medicine, Public Health and Biotechnology: Building a Man in the Machine - Proceedings of the First World Congress Part II, M. Witten (Eds.), pp. 887-901, 1996. Teaneck, NJ: World Scientific.

@incollection{sirosh:compmed96, title={A Neural Network Model of Topographic Reorganization Following Cortical Lesions}, author={Joseph Sirosh and Risto Miikkulainen}, booktitle={Computational Medicine, Public Health and Biotechnology: Building a Man in the Machine - Proceedings of the First World Congress Part II}, month={ }, editor={M. Witten}, publisher={Teaneck, NJ: World Scientific}, pages={887-901}, url="http://www.cs.utexas.edu/users/ai-lab?sirosh:compmed96", year={1996} }