I'm failing to see how these statements are comparable:
"can apparently discern handwritten digits with 75 per cent to 85 per cent accuracy, given 15 to 20 training samples of each number."
" That's not bad considering it takes thousands of training examples for more traditional neural networks to achieve 99 per cent accuracy."
To make a fair comparison, I'd need to see what the moth-model does after thousands of examples. Also, what does the traditional NN do after 15 to 20 training samples?