Self-supervised learning: generate labels from the data itself via pretext task

"mask": hold out one feature, predict it from the rest, per feature. "denoise": corrupt inputs with deterministic LCG noise, reconstruct. A function may be passed instead: pretext(X) returning list(Xp, yp).

Usage

morie_geron_self_supervised(X, pretext = "mask", noise = 0.1, seed = 0)

Arguments

X: Unlabeled data (n, d).
pretext: "mask", "denoise", or a function.
noise: Corruption scale for "denoise".
seed: LCG seed.

Value

list with loss, task_losses, predictions, targets, r2, pretext, estimate, n, method.

Examples

V <- c(1, 2, 3, 4, 5, 6, 7, 8)
morie_geron_self_supervised(V)
#> $loss
#> [1] 5.25
#> 
#> $task_losses
#> [1] 5.25
#> 
#> $predictions
#>      [,1]
#> [1,]  4.5
#> [2,]  4.5
#> [3,]  4.5
#> [4,]  4.5
#> [5,]  4.5
#> [6,]  4.5
#> [7,]  4.5
#> [8,]  4.5
#> 
#> $targets
#>      [,1]
#> [1,]    1
#> [2,]    2
#> [3,]    3
#> [4,]    4
#> [5,]    5
#> [6,]    6
#> [7,]    7
#> [8,]    8
#> 
#> $r2
#> [1] 0
#> 
#> $pretext
#> [1] "mask"
#> 
#> $estimate
#> [1] 5.25
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "Self-supervised 'mask' pretext solved by least squares; no external labels used"
#>