In cognitive science, an approach that proposes to model human information processing in terms of a network of interconnected units operating in parallel. The units are typically classified as input units, hidden units, or output units. Each unit has a default activation level that can vary as a function of the strength of (1) the inputs it receives from other units, (2) the different weights associated with its connections to the other units, and (3) its own bias. Connectionism, unlike traditional computational models in cognitive science, holds that information is distributed throughout entire networks instead of being localized in functionally discrete, semantically interpretable states