| Popis: |
We use a nonlinear neuronal network with decay and recovery of attractor eigenvalues, introduced by Kawamoto and Anderson. It shows oscillatory behavior, but delta-peaked switching time distributions. To match our experimental data, obtained with the Necker Cube as stimulus, noise has to be added. We discuss two possible levels of noise, mesoscopic or Langevin type and microscopic or physiological noise. Simulations prove that the system under Langevin noise is more robust and generates distributions best fitting our data. Microscopic noise destabilises the basins of attraction and induces a random walk leading away from stable states. We conclude that the level of noise modelling is crucial in simulations of visual perception. |
