Class LeakyRelu

java.lang.Object
deepnetts.net.layers.activation.LeakyRelu
All Implemented Interfaces:
ActivationFunction, Serializable, Consumer<Tensor>
public final class LeakyRelu extends Object implements ActivationFunction, Serializable
Leaky Rectified Linear Activation and its Derivative. y = x for x > 0, 0.1 * x for xinvalid input: '<'0 - | 1, x > 0 y' = invalid input: '<' | 0.01 , xinvalid input: '<'=0 - allow a small, positive gradient when the unit is not active https://ai.stanford.edu/~amaas/papers/relu_hybrid_icml2013_final.pdf
Author:
Zoran Sevarac
See Also:

  • Constructor Details

    • LeakyRelu

      public LeakyRelu()

    • LeakyRelu

      public LeakyRelu(float a)

  • Method Details

    • getValue

      public float getValue(float x)
      Description copied from interface: ActivationFunction
      Returns the value of activation function for specified input x
      Specified by:
      getValue in interface ActivationFunction
      Parameters:
      x - input for activation
      Returns:
      value of activation function

    • getPrime

      public float getPrime(float y)
      Description copied from interface: ActivationFunction
      Returns the first derivative of activation function for specified output y
      Specified by:
      getPrime in interface ActivationFunction
      Parameters:
      y - output of activation function
      Returns:
      first derivative of activation function

    • apply

      public void apply(Tensor3D tensor, int channel)
      Specified by:
      apply in interface ActivationFunction

    • apply

      public void apply(Tensor4D tensor, int batchIdx)
      Specified by:
      apply in interface ActivationFunction

    • apply

      public void apply(Tensor tensor, int from, int to)
      Specified by:
      apply in interface ActivationFunction