Nonlinear analysis of spatio-temporal receptive fields: IV. Generic tuning properties for rectifying rate-functions

Abstract

Neural field models have been used to explain tuning properties of V1 neurons. I here show that these models generically lead to contrast-independent tuning if they only have semilinear, zero-threshold rate functions. Other model details are unimportant, as the number and dimension of layers, imprinted feature maps, the precise form of input profiles or connectivity kernels, etc. Furthermore, also spatio-temporal solutions scale linearly with input strength but are otherwise form-invariant.