Since the model has shown that it organizes like the cortex, one can examine in more detail what the goal of this organization might be. Field (1987, 1994) and Barlow (1972) have suggested that the RFs in the primary visual cortex act as filters that form a sparse coding of the visual input. Sparse codes minimize the number of active neurons in the cortex, and are well suited for the detection of suspicious coincidences, pattern recognition, associative memory and feature grouping (Field 1994, Barlow 1972, 1985, 1990). Prior work has demonstrated that the coding of visual input produced by the RF-LISSOM model is sparse in this sense (Miikkulainen et al. 1997; Sirosh and Miikkulainen 1996, 1997; Sirosh et al. 1996). The self-organized lateral connections have proven to be crucial for reducing redundancies to achieve this coding.
By Hebbian self-organization, the lateral connections in the model learn correlations between the feature-selective cells. The stronger the correlation between two cells' activity has been in the past, the larger the connection strength between them. Because the long-range connections are inhibitory, strongly correlated regions of the network inhibit each other. At the same time, the short-range lateral excitation locally amplifies the responses of active neurons. As will be seen in chapter 4, the recurrent excitation and inhibition focuses the activity to the neurons best tuned to the features of the input stimulus, thereby producing a sparse coding of the input. This same process manifests itself in the model as tilt illusions and aftereffects, as will be shown in chapter 5.