MiS Preprint Repository

Neural field models have been used widely to explain tuning properties of cortical neurons. This paper shows that such models generically reveal tuning independent of input intensity, i.e. contrast in the visual domain, if they only have semilinear rate functions with threshold zero. Other model details are unimportant as long as spatially localized rate profiles representing receptive fields still exist. Those details concern the number and dimension of layers, and the precise form of the spatial input profiles and connectivity kernels. Arbitrary feature maps can further be imposed onto the model, and the cells in some of the model layers may reveal cell-adaptation (or facilitation) described by a passive linear kinetics. In any case, do all spatio-temporal solutions of the model scale linearly with input strength, but are otherwise form invariant. As a consequence the tuning of simple cells for orientation and spatial frequency as well as direction tuning indexes are contrast independent in these models. It is demonstrated that the results shown in the present paper apply - at least in approximation - to most models for contrast independent tuning of V1 simple cells proposed in the literature. The assumption of rectifying rate functions is discussed in the context of previous theoretical and experimental results concerning the approximate linearization of neuronal rate functions by noise.