Fig. 3.

Sample results of network testing: a the values of the error δ(j) for j randomisations (j∈(1,12000) and s = 150) and the features in the training vector w(3), w(4) and w(5). Each randomisation and calculation of the error are performed after 1000 iterations of network training; b the absolute difference between the power of the cornea predicted by the neural network and the reference power (ground truth) for 59 test cases