Figure 1.

(A) An ANN for measuring distances between pairs of pitch-classes. Stimuli are presented by activating two input units; 1 A illustrates presenting C and G. The ANN learns to activate the output units that measure (as in (B)) the distances between the two pitch-classes by processing signals from four hidden units. All output and hidden units employ a Gaussian activation function f(net) = e^{−π(net-µ)2} that ranges between 0 and 1; net is the processor’s incoming signal, and µ is the function’s mean. Weights are modified by a variant of the generalized delta rule⁹. The bar plots (C-F) depict the weights of connections between input units and each hidden unit after training ends.