Learning Rate Dropout

https://arxiv.org/abs/1912.00144

examples
{
  "class": "function:hypergan.train_hooks.learning_rate_dropout_train_hook.LearningRateDropoutTrainHook",
  "dropout": 0.01,
  "ones": 1e12,
  "zeros": 0.0,
  "skip_d": true
}
options
attribute
description
type
dropout
0-1 dropout ratio. Defaults to 0.5
float
ones
The gradient multiplier when not dropped out. Defaults to 0.1
float
zeros
The gradient multiplier when dropped out. Defaults to 0.0
float
skip_d
skip d gradients
Defaults to false
boolean
skip_g
skip g gradients
Defaults to false
boolean
Floats are configurable parameters​

JARE

Gradient Penalty

Last updated 8 months ago