A natural logarithm appears inside log loss, an error function for continuous perceptrons. Its role is to turn probability quality into a useful loss value. section 6.4
Log loss uses the probability assigned to the correct label: high probabilities produce small losses, while low probabilities produce large losses. section 6.4
A small numerical comparison
If the correct-label probability is 0.731, the loss is −ln(0.731) = 0.313. If it is 0.119, the loss is −ln(0.119) = 2.127.
