Feedback Associative Memory Networks

The following figure shows a general configuration for a feedback neural model.

A feedback network needs many iteracions before retireven the final pattern. The most popular feedback auto-association network is the Hopfield model, which has the following characteristics:

  1. Synaptic weights (predetermined by the Hebbian rule or an energy function) are prestored.
  2. Nonlinear thresholding operations are used in each stage to produce binary-valued states.
  3. State feedbacks are used so that states can be iteratively updated.
  4. Iteractions will converge to a solution that minimizes an energy function pertaining to the newtwork.
In the follwing, a sequential (asynchronous) Hopfield model and a parallel (synchronous) Hopfield model will be introduced.

Sequential (asynchronous) Hopfield model

Contents

Artificial Neural Networks
About this Tutorial