Self-organization of topographic bilinear networks for invariant recognition
Neural Computation
Hi-index | 0.01 |
We study the shape of patterns formed under different values of a control parameter in a model system for the ontogenesis of retinotopy proposed by Häussler and von der Malsburg. Guided by linear modes, their eigenvalues and nonlinear interactions, a few deciding values of the synaptic formation rate 茂戮驴are chosen, under which final states are obtained using computer simulations of the full dynamics. We find that a precise topographic mapping can only be developed under a very narrow range of 茂戮驴close to its critical value. The allowed range of 茂戮驴is relaxed if the system is equipped with a proper structure, presumably by evolution.