Computer simulation of biological inter-activation of neurons. A common simple structure involves an input layer of nodes, a middle or ‘hidden’ layer, and an output layer. Cognitive tasks are solved by progressive change in the pattern of activation over the network of interconnections. The methodology and related theory is referred to as connectionism. It differs in approach to artificial intelligence (AI). The origins of AI involve applying conventional serial processing techniques to high-level cognitive processing such as concept-formation, semantics, and symbolic processing, and it is thus based on a top-down approach. In contrast, neural nets are designed for a bottom-up approach to basic abilities of the brain such as the ability to continue functioning with ‘noisy’ and/or incomplete information, and for adaptability to changing environments by learning. Neural nets attempt to mimic and exploit the parallel processing potential of the human brain in order to deal with the sorts of problems that the brain itself is well adapted for.
See Activation (in a connectionist model), Artificial intelligence (AI), Auto-encoder networks, Backpropagation, Cognitive neuroscience, Cognitive science, Computational models, Connectionism, Connectionist models, Distributed representation, Genetic algorithms, Neuroconstructivist networks, Non-linear associator, On-line emergence