File:Fusing-Swarm-Intelligence-and-Self-Assembly-for-Optimizing-Echo-State-Networks-642429.f1.ogv

English: This animation is an example of the network growth process governed by the SINOSA model. The small colored spheres are growth cones, which occur at the leading tips of growing connections (axons) and are responsible for guiding the connections from emitting to target cells (neurons). The large spheres represent the regions of space around cells that growth cones must enter in order for a connection to be established between a growth cone's emitting cell and the corresponding target cell. The weight on a connection established by a growth cone depends on the current position of the growth cone with respect to the center of the cell's “sphere-of-influence”. For clarity, only a small fraction of the total number of growth cones involved in the growth process are shown and the connections are not shown. There are seven different sets of growth cones shown, which are indicated by color. In this example each growth cone has a single target cell that it is allowed to establish connections with. The growth cones within a set all have the same target cell, and it is exactly these growth cones that interact with one another via a mechanism inspired by particle swarm optimization. Growth occurs in a three-dimensional space and the movements of the growth cones are not bounded. The growth process is initialized by placing each growth cone at a randomly selected location within the sphere-of-influence of its single target cell. Initially, the growth cones exhibit highly variable, wide-ranging movements as they explore different weighted connections or no connection at all. As the growth process progresses, and better networks are discovered, the growth cones exploit this information and begin to converge on single points in space, which ultimately represents a single best-performing network.