Package org.deeplearning4j.nn.weights

Interface Summary

Interface	Description
IWeightInit	Interface for weight initialization.

Class Summary

Class	Description
WeightInitConstant	Initialize to a constant value (deafult 0).
WeightInitDistribution	Sample weights from a provided Distribution Note that Distribution is not extendable as it is interpreted through Distributions.createDistribution(Distribution).
WeightInitIdentity	Weights are set to an identity matrix.
WeightInitLecunUniform	Uniform U[-a,a] with a=3/sqrt(fanIn).
WeightInitNormal	Normal/Gaussian distribution, with mean 0 and standard deviation 1/sqrt(fanIn).
WeightInitRelu	: He et al.
WeightInitReluUniform	He et al.
WeightInitSigmoidUniform	A version of WeightInitXavierUniform for sigmoid activation functions.
WeightInitUniform	Uniform U[-a,a] with a=1/sqrt(fanIn).
WeightInitUtil	Weight initialization utility
WeightInitVarScalingNormalFanAvg	Truncated aussian distribution with mean 0, variance 1.0/((fanIn + fanOut)/2)
WeightInitVarScalingNormalFanIn	Gaussian distribution with mean 0, variance 1.0/(fanIn) If a scale is provided, use variance scale/(fanIn) instead
WeightInitVarScalingNormalFanOut	Truncated normal distribution with mean 0, variance 1.0/(fanOut) If a scale is provided, variance is scale / fanOut
WeightInitVarScalingUniformFanAvg	Uniform U[-a,a] with a=3.0/((fanIn + fanOut)/2)
WeightInitVarScalingUniformFanIn	Uniform U[-a,a] with a=3.0/(fanIn) If a scale is provided, a = 3.0 * scale / (fanIn)
WeightInitVarScalingUniformFanOut	Uniform U[-a,a] with a=3.0/(fanOut) If a scale is provided, a = 3.0 * scale / fanOut
WeightInitXavier	As per Glorot and Bengio 2010: Gaussian distribution with mean 0, variance 2.0/(fanIn + fanOut)
WeightInitXavierLegacy	Xavier weight init in DL4J up to 0.6.0.
WeightInitXavierUniform	As per Glorot and Bengio 2010: Uniform distribution U(-s,s) with s = sqrt(6/(fanIn + fanOut))

Enum Summary

Enum	Description
WeightInit	Weight initialization scheme