SUBSCRIPTIONS & PRICING
GENERAL INFORMATION
chapter 9, Recurrent Neural Networks
Table of Contents
- 1. Introduction
Chapter Contents
- 9.1. Hopfield Neural Networks
- 9.2. The Bidirectional Associative Memory (BAM)
- 9.3. The Generalized Linear Neural Network
- 9.3.1. GLNN Example
- 9.4. Real-Time Recurrent Network
- 9.5. Elman Recurrent Network
Excerpt
Recurrent neural networks are networks that feed the outputs from neurons to other adjacent neurons, to themselves, or to neurons on preceding network layers. Two of the most popular recurrent neural networks are the Hopfield and the Bidirectional Associative Memory (BAM) networks.
9.1 Hopfield Neural Networks
The Hopfield network [Hopfield, 1982] rekindled interest in neural networks in the early 1980s, but it is rarely used today. The Hopfield network is often used as an auto-associative memory or content-associated network with fully recurrent connections between the input and output, as depicted in Fig. 9.1. Its primary purpose is to retrieve stored patterns by presenting a portion of the desired pattern to the network.
In the Hopfield neural network, each neuron is connected to every other neuron through weighted synaptic links. A set of P patterns with bits encoded as Mpi∈{−1, 1} are encoded in the synaptic weights by using an outer-product learning rule [Hopfield, 1982]. The encoded synaptic weights are determined by

©2005 Society of Photo-Optical Instrumentation Engineers











This Publication
Google Scholar
PubMed