|
A neural network is a collection of multiple independent processors configured such that each processor is in direct communication with all adjacent processors. These connections form a network of “neurons,” much like the structure of the human brain. Neural networks are being studied by artificial intelligence researchers because of their ability to learn spontaneously.
When a neural network is presented with a pattern (such as a visual image on the retina of the eye) that can be mapped onto the matrix of processors (such as the rods and cones of the retina), each processor responds independently and may or may not be activated depending on the pattern presented. Since each processor is in constant communication with the processors adjacent to it, each adjusts its activation threshold and communication tolerances with adjacent processors until a state of equilibrium is reached throughout the network. The pattern is thus distributed and “stored” network-wide in this set of thresholds and tolerances.
If only a portion of the pattern is presented later, the processors that are activated will communicate with adjacent processors until the previously attained equilibrium has been reestablished. The result is the “recognition” of the complete pattern from only a small fragment.