Activations

Softmax

layer eddl::Softmax(layer parent, int axis = -1, string name = "")

Applies a Jacobian Softmax activation function to the given layer.

See

https://en.wikipedia.org/wiki/Softmax_function

Parameters
  • parent – Parent layer

  • axis – Dimension in which to operate. Default -1, which uses the last axis

  • name – Name of the layer

Returns

Output of Softmax transformation

The Softmax activation function is: softmax(x) = exp(x) / reduce_sum(exp(x))

Example:

l = Softmax(l);

Sigmoid

layer eddl::Sigmoid(layer parent, string name = "")

Applies a Sigmoid activation function to the given layer.

See

https://en.wikipedia.org/wiki/Sigmoid_function

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of Sigmoid activation

The Sigmoid activation function is: sigmoid(x) = 1 / (1 + exp(-x))

Example:

l = Sigmoid(l);

ReLu

layer eddl::ReLu(layer parent, string name = "")

Applies a Rectified Linear Unit activation function to the given layer.

See

https://en.wikipedia.org/wiki/Rectifier_(neural_networks)

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of ReLu activation

The ReLu activation function is:

  • if x > 0: relu(x) = x

  • else: relu(x) = 0

Example:

l = ReLu(l);

Thresholded ReLu

layer eddl::ThresholdedReLu(layer parent, float alpha = 1.0, string name = "")

Applies the Thresholded version of a Rectified Linear Unit activation function to the given layer.

Parameters
  • parent – Parent layer

  • alpha – Threshold value

  • name – Name of the layer

Returns

Output of Thresholded ReLu activation

The Thresholded ReLu activation function is:

  • if x > alpha: threshdolded_relu(x) = x

  • else: thresholded_relu(x) = 0

Example:

l = ThresholdedReLu(l, 1.0);

Leaky ReLu

layer eddl::LeakyReLu(layer parent, float alpha = 0.01, string name = "")

Applies the Leaky version of a Rectified Linear Unit activation function to the given layer.

See

https://en.wikipedia.org/wiki/Rectifier_(neural_networks)#Leaky_ReLUs

Parameters
  • parent – Parent layer

  • alpha – Negative slope coefficient

  • name – Name of the layer

Returns

Output of Leaky ReLu activation

The Leaky ReLu activation function is:

  • if x > 0: leaky_relu(x) = x

  • else: leaky_relu(x) = alpha * x

Example:

l = LeakyReLu(l, 0.01);

ELu

layer eddl::Elu(layer parent, float alpha = 1.0, string name = "")

Applies the Exponential Linear Unit activation function to the given layer.

Parameters
  • parent – Parent layer

  • alpha – ELu coefficient

  • name – Name of the layer

Returns

Output of ELu activation

The ELu activation function is:

  • if x > 0: elu(x) = x

  • else: elu(x) = alpha *  (exp(x) - 1)

Example:

l = Elu(l, 1.0);

SeLu

layer eddl::Selu(layer parent, string name = "")

Applies the Scaled Exponential Linear Unit activation function to the given layer.

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of Selu activation

The SeLu activation function is:

  • if x > 0: selu(x) = scale * x

  • else: selu(x) = scale * (alpha *  (exp(x) - 1))

where alpha = 1.6732632423543772848170429916717 and scale = 1.0507009873554804934193349852946

Example:

l = Selu(l);

Exponential

layer eddl::Exponential(layer parent, string name = "")

Applies the Exponential (base e) activation function to the given layer.

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of Exponential activation

The Exponential activation function is: exp(x)

Example:

l = Exponential(l);

Softplus

layer eddl::Softplus(layer parent, string name = "")

Applies the Softplus activation function to the given layer.

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of Exponential activation

The Softplus activation function is: softplus(x) = log(1 + exp(x))

Example:

l = Softplus(l);

Softsign

layer eddl::Softsign(layer parent, string name = "")

Applies the Softsign activation function to the given layer.

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of Exponential activation

The Softsign activation function is: softsign(x) = x / (1 + abs(x))

Example:

l = Softsign(l);

Linear

layer eddl::Linear(layer parent, float alpha = 1.0, string name = "")

Applies the Linear activation function to the given layer.

Parameters
  • parent – Parent layer

  • alpha – Linear coefficient

  • name – Name of the layer

Returns

Output of Linear activation

The Linear activation function is: linear(x) = alpha * x

Example:

l = Linear(l, 1.0);

Tanh

layer eddl::Tanh(layer parent, string name = "")

Applies the Hyperbolic tangent activation function to the given layer.

See

https://en.wikipedia.org/wiki/Hyperbolic_function

Parameters
  • parent – Parent layer

  • name – Name of the layer

Returns

Output of hyperbolic activation

The Tanh activation function is: tanh(x) = sinh(x)/cosh(x) = ((exp(x) - exp(-x))/(exp(x) + exp(-x)))

Example:

l = Tanh(l);