1. a technique for modeling the neural changes in the brain that underlie cognition and perception in which a large number of simple hypothetical neural units are connected to one another.

2. an artificial intelligence system used for learning and classifying data and applied in research on pattern recognition, speech recognition, machine translation of languages, and financial prediction, among other areas. Neural networks are usually abstract structures modeled on a computer and consist of a number of interconnected processing elements (nodes), each with a finite number of inputs and outputs. The elements in a network can have a “weight” determining how they process data, which can be adjusted according to experience. In this way, the network can be trained to recognize patterns in input data by optimizing the output of the network. The analogy is with the supposed action of neurons in the brain. In addition, neural networks are often structured into layers, including an input layer (in which properties of input parameters are encoded), possibly multiple hidden layers (in which generalizations of the input parameters are reflected), and an output layer (in which the response of the neural network system is reported to the environment). The connectivity of these layers often differs, usually reflecting the algorithms the neural network uses for learning. There are multiple families of algorithms used for learning patterns in data, including Hebbian learning and back-propagation learning. See also perceptron.