Computational Maps in the Visual Cortex
     Figure 5.32
MiikkulainenBednarChoeSirosh
Home    
About the Authors
Back Cover    
Table of Contents 
Sample Chapter 
Figures    
References    
Errata    
Demos     
Talks/Courses 
Software    
Credits    
Purchase online at:

springeronline.com
amazon.com

Click on the image to see a PDF version (for zooming in)

Fig. 5.32. Effect of training patterns on OR/OD/DR maps. From left to right, each row shows a sample retinal activation, final spatiotemporal receptive fields of sample neurons with lags 3, 2, 1, and 0 from left to right, their inhibitory lateral connections, the orientation preference and selectivity map, the ocular dominance map, and the direction preference and selectivity map. The top row shows a network trained with oriented Gaussians, the middle with noisy disks, and the bottom with natural images. All networks develop realistic orientation maps. Maps with Gaussian inputs and noisy disks develop only smooth, two- or three-lobed RFs, whereas networks with natural images develop a wide variety of RF types, corresponding to the wide range of patterns seen in natural images. In each case, the lateral connection patterns follow the features in the map. These results show that joint OR/OD/DR maps can develop from a variety of abstract and realistic input stimuli.