search

weightagnostic / weightagnostic.github.io

repo for interactive article
https://arxiv.org/abs/1906.04358
Creative Commons Attribution 4.0 International
456 stars 110 forks source link

A question about the Test Accuracy in MNIST experiment #9

Open 3riccc opened 5 years ago

3riccc commented 5 years ago

Hi, I want to ask a question about the test accuracy. In WANN with Random Weight, the test accuracy is 82.0% ± 18.7%. but 82% + 18.7% = 100.7% which is greater than 100%. so what does this mean?

weightagnostic commented 5 years ago

Hi there,

Please see this response from a related discussion.

82.0% ± 18.7% is the mean and standard deviation not the min/max (if we got 100.7% on MNIST we definitely would have led with that).

You can tell by the individual digit accuracy that the distribution isn't normal so it ends up a bit wonky...

Thanks