Neuronal tension may co-shape V1 orientation maps

The structure of orientation maps, has been shown to minimize the length of horizontal connections in V1, given certain connection patterns as a function of orientation difference. We take a V1 model network with horizontal connections. Neural Activations are maintained in this network by recurrent computations constituting an associator network. Weight Learning has been performed for the purpose of memorizing the network’s internal representation of natural image patches. Neuronal Shifting is performed here to assess whether minimizing the lengths of the learnt connections leads to a realistic orientation map. After convergence, horizontally directed tension forces are in balance. The results with 1024 neurons and 16×16 pixel retinal input show that the neurons arrange topographically and form an orientation map similar to one hypercolumn in V1.

@InProceedings{WT07b, 
 	 author =  {Weber, Cornelius and Triesch, Jochen},  
 	 title = {Neuronal tension may co-shape V1 orientation maps}, 
 	 booktitle = {Proc. 31st Göttingen Neurobiology Conference},
 	 editors = {},
 	 number = {},
 	 volume = {},
 	 pages = {},
 	 year = {2007},
 	 month = {},
 	 publisher = {},
 	 doi = {}, 
 }