Recurrent Neural Network Program

This option involves choosing one of three network architectures: creating a Hopfield Network, or a Boltzmann Machine or a Restricted Boltzmann Machine. These are described below.

Hopfield Networks:

For this assignment, implement the Hopfield Network that is based on 8 exemplars, each 100 elements in length.

These exemplars are to be displayed in a 10x10 rasterized array and should depict the first 8 numerals: 0 – 7.

The output of this program should display the sequence of iterates displayed in the 10x10 array in similar fashion as the examples shown in Lippman’s paper “Introduction to Neural Networks”. Display at least two sequences each sequence starting with a noisy version of one of the exemplars. Feel free to experiment.

Question: Where is the Lippman's paper and the examples?

The result of the network:

V1 Input

1. Training 2 numbers works well
2. Training more than 3 numbers does not really work despite pattern does not overlap