Relating real-time backpropagation and backpropagation-through-time: an application of flow graph interreciprocity

Authors:
Françoise Beaufays;Eric A. Wan
Affiliations:
-;-
Venue:
Neural Computation
Year:
1994

Citing 0
Cited 4

Evenet 2000: Designing and Training Arbitrary Neural Networks in Java

IWANN '01 Proceedings of the 6th International Work-Conference on Artificial and Natural Neural Networks: Bio-inspired Applications of Connectionism-Part II
A Signal-Flow-Graph Approach to On-line Gradient Calculation

Neural Computation
Diagrammatic derivation of gradient algorithms for neural networks

Neural Computation
Convergence of online gradient method for feedforward neural networks with smoothing L 1/2 regularization penalty

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

We show that signal flow graph theory provides a simple way torelate two popular algorithms used for adapting dynamic neuralnetworks, real-time backpropagation andbackpropagation-through-time. Starting with the flow graph forreal-time backpropagation, we use a simple transposition to producea second graph. The new graph is shown to be interreciprocal withthe original and to correspond to the backpropagation-through-timealgorithm. Interreciprocity provides a theoretical argument toverify that both flow graphs implement the same overall weightupdate.