Stack filters and neural networks
Edward J. Coyle, Neal C. Gallagher · 2003
The stack filter approach, which provides a unique interpretation of the function of each neuron in the network when the goal is to minimize the mean absolute error, is described. Stack filters also provide information on when soft decisions, or sigmoid functions, are necessary for the neural network to attain optimality. The associative memory behavior exhibited by some stack filters is also reviewed and compared with previous approaches to associate memory. An image processing example is provided to demonstrate the use of a new learning algorithm for stack filters.>