A statistical neural field approach to orientation selectivity

We apply the recently proposed statistical neural eld approach 4,2] for modeling orientation selectivity in the primary visual cortex. Firstly, we demonstrate that the neural eld approach is a powerful tool for modeling neural structures with speciic lateral connections. Secondly, we test in a biologically more plausible way our hypothesis 1] that orientation bias and tuning in macaque striate cortex can be generated by the same lateral interactions. The spiking neural model shows that (i) contrast invariant tuning emerges, and (ii) the tuning dynamics of the membrane potential and the ring rate are in accordance with observations which until now seemed to be contradictory 3,5,7,6]. 1 The statistical model Following 9,4] we describe the population activity of neurons as well as the propagation of spikes in terms of probability density functions (p.d.f.). For a neuron of type s at each spatial location r in the neural eld the probability density g s (r; u; t) for being at a subthreshold membrane potential u at time t is described via a single compartmental model. When a cell res, it emits as many \spikes" as it has synaptic terminals. The probability density f s (r; t) for spikes emitted by neuron of type s being at a position r in the neural eld at time t traveling at direction is then determined as following. The spikes propagate radially in all directions and diiuse in space simulating the eeect of diierent propagation delays of action potentials. Synaptic connections are modeled by absorption of spikes at each point in the neural