Open main menu

Wikipedia β

In machine learning, the Delta rule is a gradient descent learning rule for updating the weights of the inputs to artificial neurons in a single-layer neural network.[1] It is a special case of the more general backpropagation algorithm. For a neuron with activation function , the delta rule for 's th weight is given by

,

where

is a small constant called learning rate
is the neuron's activation function
is the target output
is the weighted sum of the neuron's inputs
is the actual output
is the th input.

It holds that and .

The delta rule is commonly stated in simplified form for a neuron with a linear activation function as

While the delta rule is similar to the perceptron's update rule, the derivation is different. The perceptron uses the Heaviside step function as the activation function , and that means that does not exist at zero, and is equal to zero elsewhere, which makes the direct application of the delta rule impossible.

Derivation of the delta ruleEdit

The delta rule is derived by attempting to minimize the error in the output of the neural network through gradient descent. The error for a neural network with   outputs can be measured as

 .

In this case, we wish to move through "weight space" of the neuron (the space of all possible values of all of the neuron's weights) in proportion to the gradient of the error function with respect to each weight. In order to do that, we calculate the partial derivative of the error with respect to each weight. For the  th weight, this derivative can be written as

 .

Because we are only concerning ourselves with the  th neuron, we can substitute the error formula above while omitting the summation:

 

Next we use the chain rule to split this into two derivatives:

 

To find the left derivative, we simply apply the general power rule:

 

To find the right derivative, we again apply the chain rule, this time differentiating with respect to the total input to  ,  :

 

Note that the output of the  th neuron,  , is just the neuron's activation function   applied to the neuron's input  . We can therefore write the derivative of   with respect to   simply as  's first derivative:

 

Next we rewrite   in the last term as the sum over all   weights of each weight   times its corresponding input  :

 

Because we are only concerned with the  th weight, the only term of the summation that is relevant is  . Clearly,

 ,

giving us our final equation for the gradient:

 

As noted above, gradient descent tells us that our change for each weight should be proportional to the gradient. Choosing a proportionality constant   and eliminating the minus sign to enable us to move the weight in the negative direction of the gradient to minimize error, we arrive at our target equation:

 .

See alsoEdit

ReferencesEdit

  1. ^ Russell, Ingrid. "The Delta Rule". University of Hartford. Archived from the original on 4 March 2016. Retrieved 5 November 2012.