Denoising autoencoder loss against the clean input (Geron Ch 18, grdae)

Scoring against x rather than x_tilde is the point: copying the input is now a losing strategy. denoising_gain = noise_energy / loss.

Usage

morie_geron_denoising_autoencoder(x, noise, decoded, corruption = "additive")

Arguments

x

Clean inputs (m, d).

noise

Corruption, broadcastable, or a 0/1 keep-mask when corruption = "dropout".

decoded

Reconstructions.

corruption

"additive" or "dropout".

Value

List with loss, mse_per_element, per_sample_loss, x_tilde, noise_energy, denoising_gain, snr_db.

Examples

morie_geron_denoising_autoencoder(x = c(1, 2, 3, 4, 5, 6, 7, 8),
  noise = c(1, 2, 3, 4, 5, 6, 7, 8), decoded = c(1, 2, 3, 4, 5, 6, 7, 8))
#> $loss
#> [1] 0
#> 
#> $mse_per_element
#> [1] 0
#> 
#> $per_sample_loss
#> [1] 0 0 0 0 0 0 0 0
#> 
#> $x_tilde
#>      [,1]
#> [1,]    2
#> [2,]    4
#> [3,]    6
#> [4,]    8
#> [5,]   10
#> [6,]   12
#> [7,]   14
#> [8,]   16
#> 
#> $noise_energy
#> [1] 25.5
#> 
#> $denoising_gain
#> [1] Inf
#> 
#> $snr_db
#> [1] 0
#> 
#> $corruption
#> [1] "additive"
#> 
#> $estimate
#> [1] 0
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "Denoising autoencoder reconstruction loss"
#>