Basic..::..BatchNorm Method (Variable, Single, String, String, String, String, Boolean, Single, Single)
Batch normalization layer (Ioffe and Szegedy, 2014). Normalize the activations of the previous layer at each batch, i.e.applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1.
Namespace:
SiaNet.NNAssembly: SiaNet (in SiaNet.dll)
Syntax
public static Function BatchNorm( Variable layer, float epsilon, string betaInitializer, string gammaInitializers, string runningMeanInitializer, string runningStdInvInitializer, bool spatial, float normalizationTimeConstant, float blendTimeConst )
Public Shared Function BatchNorm ( _ layer As Variable, _ epsilon As Single, _ betaInitializer As String, _ gammaInitializers As String, _ runningMeanInitializer As String, _ runningStdInvInitializer As String, _ spatial As Boolean, _ normalizationTimeConstant As Single, _ blendTimeConst As Single _ ) As Function
Parameters
- layer
- Type: Variable
The output of the last layer.
- epsilon
- Type: Single
Small float added to variance to avoid dividing by zero.
- betaInitializer
- Type: String
Initializer for the beta weight.
- gammaInitializers
- Type: String
Initializer for the gamma weight.
- runningMeanInitializer
- Type: String
Initializer for the running mean weight.
- runningStdInvInitializer
- Type: String
Initializer for the running standard inv weight.
- spatial
- Type: Boolean
Boolean, if yes the input data is spatial (2D). If not, then sets to 1D
- normalizationTimeConstant
- Type: Single
The time constant in samples of the first-order low-pass filter that is used to compute mean/variance statistics for use in inference
- blendTimeConst
- Type: Single
The blend time constant in samples.