AdaDelta

public class AdaDelta

Optimizer that implements the Adadelta algorithm.

Adadelta optimization is a stochastic gradient descent method that is based on adaptive learning rate per dimension to address two drawbacks:

  • the continual decay of learning rates throughout training
  • the need for a manually selected global learning rate

Adadelta is a more robust extension of Adagrad that adapts learning rates based on a moving window of gradient updates, instead of accumulating all past gradients. This way, Adadelta continues learning even when many updates have been done. Compared to Adagrad, in the original version of Adadelta you don't have to set an initial learning rate. In this version, initial learning rate can be set, as in most other optimizers.

According to section 4.3 ("Effective Learning rates"), near the end of training step sizes converge to 1 which is effectively a high learning rate which would cause divergence. This occurs only near the end of the training as gradients and step sizes are small, and the epsilon constant in the numerator and denominator dominate past gradients and parameter updates which converge the learning rate to 1.

According to section 4.4("Speech Data"),where a large neural network with 4 hidden layers was trained on a corpus of US English data, ADADELTA was used with 100 network replicas.The epsilon used is 1e-6 with rho=0.95 which converged faster than ADAGRAD, by the following construction: new AdaDelta(graph, 1.0f, 0.95f, 1e-6f);

Constants

String ACCUMULATOR
String ACCUMULATOR_UPDATE
float EPSILON_DEFAULT
float LEARNING_RATE_DEFAULT
float RHO_DEFAULT

Inherited Constants

Public Constructors

AdaDelta ( Graph graph, float learningRate)
Creates an AdaDelta Optimizer
AdaDelta ( Graph graph, float learningRate, float rho, float epsilon)
Creates an AdaDelta Optimizer
AdaDelta ( Graph graph, String name, float learningRate)
Creates an AdaDelta Optimizer
AdaDelta ( Graph graph, String name, float learningRate, float rho, float epsilon)
Creates an AdaDelta Optimizer

Public Methods

String
getOptimizerName ()
Get the Name of the optimizer.
String

Inherited Methods

Constants

public static final String ACCUMULATOR

Constant Value: "accum"

public static final String ACCUMULATOR_UPDATE

Constant Value: "accum_update"

public static final float EPSILON_DEFAULT

Constant Value: 1.0E-7

public static final float LEARNING_RATE_DEFAULT

Constant Value: 0.001

public static final float RHO_DEFAULT

Constant Value: 0.95

Public Constructors

public AdaDelta ( Graph graph)

public AdaDelta ( Graph graph, float learningRate)

Creates an AdaDelta Optimizer

Parameters
graph the TensorFlow Graph
learningRate the learning rate

public AdaDelta ( Graph graph, float learningRate, float rho, float epsilon)

Creates an AdaDelta Optimizer

Parameters
graph the TensorFlow Graph
learningRate the learning rate
rho The decay factor
epsilon A constant epsilon used to better conditioning the grad update

public AdaDelta ( Graph graph, String name, float learningRate)

Creates an AdaDelta Optimizer

Parameters
graph the TensorFlow Graph
name the name for this Optimizer (defaults to 'Adadelta')
learningRate the learning rate

public AdaDelta ( Graph graph, String name, float learningRate, float rho, float epsilon)

Creates an AdaDelta Optimizer

Parameters
graph the TensorFlow Graph
name the name for this Optimizer (defaults to 'Adadelta')
learningRate the learning rate
rho The decay factor
epsilon A constant epsilon used to better conditioning the grad update

Public Methods

public String getOptimizerName ()

Get the Name of the optimizer.

Returns
  • The optimizer name.

public String toString ()