Fig. 9

Embedded LeNet-5 training summary for the lateral crania dataset. Note the greater length of the training cycle relative to the dorso-ventral view analysis (see Fig. 7), the steady improvement in the loss ratio during the whole of the training interval, and the distinct reduction in loss improvement as the point of convergence is reached at or around 36 training rounds. The entire training cycle took 35 s of CPU time to complete