Author: Elman, J.
Title: Finding Structure in Time
Publication: CRL Technical Report 8801
Date: 1988 San Diego
This technical report describes the advantages of incorporating a temporal element in neural network processing. While simple feed-forward neural nets process each input in a relative void, Jeff Elman adds sequence or time to the equation by introducing a hidden layer that remembers the state of the network after the last input. Feeding both the new input and the “context” into the net expands its recognition capabilities. neural networks modeling cybernetics