Abstract
Unsupervised learning models have been proposed based on experience (Ahumada and Mulligan 1990; Wachtler, Doi, Lee and Sejnowski, 2007) that allow the cortex to develop units with L/M specific color opponent receptive fields like the blob cells reported by Hubel and Wiesel (), on the basis of visual experience. These models used ganglion cells with L/M indiscriminate wiring as inputs to the learning mechanism, which was presumed to occur at the cortical level. Since much research has indicated L/M specific wiring at the ganglion cell level (Reid and Shapley, 2002), an unsupervised learning model for developing L/M specific wiring at the ganglion cell level is proposed. The model takes as input the ouputs of simulated bipolar cells and, for the single cone center case, it is assumed that the surround for an on-center bipolar is restricted to off-center bipolars and vice versa. The learning mechanism needs only reduce the weight of the surround cells connected to the same type of cone. In the previous models, the ganglion cell center-surround mechanism provided the decorrelation of the L and M cones necessary for learning the distinction. In this model the decorrelation is provided by a simple model of the horizontal cells, which provides an average L+M signal that defines the on-off boundary. The possible value of this mechanism in terms of improved color discrimination is discussed.
The author was supported by NASA Space Human Factors Research Project WBS 466199 and by the NASA Engineering and Safety Center.