Chaotic neural networks for multi-resolution analysis

In this paper, we investigate new dynamic neural networks for brain data multi-resolution analysis. It is based on chaotic neuron model. Multi-resolution chaotic neural network (MRCNN) architecture is built by cascading the single-layer neural sub-networks, and a higher layer learns to cluster the prototypes developed at the layer directly below it. They have multi-output in coarse-to-fine hierarchical manner, which can reveal the inherent structural characteristic of their input data. A learning processing is also derived from training weights of the networks. They are availably applied to brain data analysis.