Skip to main content
Fig. 1 | Biology Direct

Fig. 1

From: What can ecosystems learn? Expanding evolutionary ecology with learning theory

Fig. 1

Training a Hopfield network using Hebb’s rule. Network training: Unsupervised learning processes as used to train a Hopfield network to store two configurations, patterns A and B. Each unit in the Hopfield network corresponds to a pixel in the image display. Six units are highlighted to illustrate the changes to connections during training in pattern A. Hebbs rule alters connections between units such that units of the same sign (1:1 or -1:-1) become more correlated (blue lines) and units of opposite signs (1:-1 or -1:1) become more anti-correlated (red lines). Network behaviour: Training the network on both patterns results in a network with attractors (a.k.a. memories) for these patterns and system dynamics result in all initial conditions converging to one of the trained patterns (a). This behaviour enables these systems to be used for a variety of functions, including: (b) recovery of complete composition from partial input; (c) noise reduction; and (d) classification (the input image is a closer match for the plane configuration than the bird configuration)

Back to article page