Functional Layers
These functions expose the backend of Lux.jl
. In the long-term we plan to move these into NNlib
Lux.dropout
— Functiondropout(rng::AbstractRNG, x, p, q, dims, ::Val{training})
dropout(rng::AbstractRNG, x, mask, p, q, dims, t::Val{training}, ::Val{update_mask})
If training
then dropout is applied on x
with probability p
along dims
. If mask
is passed it is used if update_mask
is false. If update_mask
is true then the mask is generated and used.
Lux.normalization
— Functionnormalization(x, running_mean, running_var, scale, bias, activation, reduce_dims,
::Val{training}, momentum, epsilon)
Performs BatchNorm/GroupNorm/InstanceNorm based on input configuration
Detailed docs are WIP