Artificial Neural Net Attractors

J. C. Sprott
Department of Physics, University of Wisconsin, Madison, WI 53706, USA


Aesthetically appealing patterns are produced by the dynamical behavior of artificial neural networks with randomly chosen connection strengths.  These feed-forward networks have a single hidden layer of neurons and a single output, which is fed back to the input to produce a scalar time series that is always bounded and often chaotic.  Sample attractors are shown and simple computer code is provided to encourage experimentation.

Ref: J. C. Sprott, Comput. & Graphics 22, 143-149 (1998)

The complete paper is available in PDF format.

Return to Sprott's Books and Publications.

Fig. 1. A feed-forward neural network with D inputs, N neurons, and a single output that is fed back to the input.
[Figure 1]

Fig. 2. Sample neural network attractors.
[Figure 2a]
[Figure 2b}
[Figure 2c]
[Figure 2d]

Fig. 3. Additional neural network attractors.
[Figure 3]

The computer source code nnet256.bas from the article is available along with an executable version nnet256.exe. Many more images of this type are available.