查看文章

Many of the chapters in Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Volume 1: Foundations make use of the ability of a parallel network to perform cooperative searches for good solutions to problems. The basic idea is simple: The weights on the connections between processing units encode knowledge about how things normally fit together in some domain and the initial states or external inputs to a subset of the units encode some fragments of a structure within the domain. These fragments constitute a problem: What is the whole structure from which they probably came? The network computes a" good solution" to the problem by repeatedly updating the states of units that represent possible other parts of the structure until the network eventually settles into a stable state of activity that represents the solution. One field in which this style of computation seems particularly ap …