Formula: mu = (1/H) sum_i a_i, sigma = sqrt((1/H) sum_i (a_i - mu)^2), y = gamma (a - mu)/sigma + beta
References
Ba, Kiros and Hinton (2016), Layer Normalization, arXiv:1607.06450, equation (3). Verified against the paper.
Examples
V <- c(1, 2, 3, 4, 5, 6, 7, 8)
Vitlnorm(V)
#> $y
#> [,1] [,2] [,3] [,4] [,5] [,6] [,7]
#> [1,] -1.527525 -1.091089 -0.6546537 -0.2182179 0.2182179 0.6546537 1.091089
#> [,8]
#> [1,] 1.527525
#>
#> $mu
#> [1] 4.5
#>
#> $sigma
#> [1] 2.291288
#>
#> $n
#> [1] 1
#>
#> $H
#> [1] 8
#>
#> $method
#> [1] "Layer normalisation"
#>
#> attr(,"class")
#> [1] "morie_rich_result" "list"
