©2020-2024 ioDraw All rights reserved
Recently, there is such a structure in the paper ：
There is one after pooling l2-norm.norm yes normalization Abbreviation for .Ok, Look what this is ：
Standardization ? Normalization ? normalization ?…
<> right key
L2 normalization ： Turn a set of numbers into 0-1 between .pytorch The function called is F.normalize. This is how the document is written ：
about L2 For ,p=2, The denominator is ( x 1 2 + x 2 2 + . . + x n 2 ) \sqrt{( x_1^2 + x_2^2 + ...
+x_n^2)}(x12+x22+...+xn2) , The molecule is x i x_i xi. So if there's one tensor It looks like this ：
20.2 20. 2 2 + 30. 3 2 = 0.5547001962252291
\frac{20.2}{\sqrt{20.2^2+30.3^2}}=0.554700196225229120.22+30.32 20.2=0.5547001
962252291
30.3 20. 2 2 + 30. 3 2 = 0.8320502943378437
\frac{30.3}{\sqrt{20.2^2+30.3^2}}=0.832050294337843720.22+30.32 30.3=0.8320502
943378437
40.4 40. 4 2 + 50. 5 2 = 0.6246950475544243
\frac{40.4}{\sqrt{40.4^2+50.5^2}}=0.624695047554424340.42+50.52 40.4=0.6246950
475544243
50.5 40. 4 2 + 50. 5 2 = 0.7808688094430304
\frac{50.5}{\sqrt{40.4^2+50.5^2}}=0.780868809443030440.42+50.52 50.5=0.7808688
094430304
Now let's test it ：
Hee hee ,666!
Technology