n-dimensional
the negative slope of the rectifier used after this layer (only used with 'leakyRelu'
)
either 'fanIn'
(default) or 'fanOut'
. Choosing 'fanIn'
preserves the magnitude of the variance of the weights in the
forward pass. Choosing 'fanOut'
preserves the magnitudes in the
backwards pass.
the non-linear function,
recommended to use only with 'relu'
or 'leakyRelu'
(default).
Generated using TypeDoc
Fills the input
Tensor
with values according to the method described inDelving deep into rectifiers: Surpassing human-level performance on ImageNet classification
- He, K. et al. (2015), using a uniform distribution.The resulting tensor will have values sampled from
U(-bound, bound)
wherebound = gain * sqrt(3/fanMode)
Also known as He initialization.
This is the default initializer for
Conv
andLinear
layers.