10 ms·
> Most networks gained a large percentage of their final accuracy in just one epoch. And it usually was the case that a higher accuracy in the first epoch meant
by scotty79 11y ago
> Most networks gained a large percentage of their final accuracy in just one epoch. And it usually was the case that a higher accuracy in the first epoch meant a higher final accuracy.
This sounds to me like learning was just crawling to local optimum not actually exploring or making any breakthrough in understanding of the domain.