Tensorflow avoid /accurate?

Hello,
at 6:39 of the chapter Improved implementation of softmax, prof Andrew said tensorflow can avoid very large or small numbers.
I can’t understand this thing.
Tensorflow is supposed to give a more accurate value.Then how does avoiding work?

Hi @farhana_hossain,

Check this out please.

https://community.deeplearning.ai/t/why-is-from-logits-true-plus-activation-linear-more-stable/431907

Cheers,
Raymond

1 Like