Anny on

Class: Connection

Neuron.Connection

Neurons communication via Connections. Their weights determine the output of the network and are updated during training. The knowledge or ability of a network is represented in the weight matrix (all the weight values).

Constructor

new Connection(source, target, weight)

Parameters:
Name Type Description
source Neuron

The Neuron that will send its output to the target Neuron.

target Neuron

The Neuron that will get its input from the source Neuron.

weight number

The strength of the connection. Meaning, what ratio of the source Neuron's output is passed to the target Neuron's input.

Source:
See:

Members

source :Neuron

A reference to the Neuron at the start of this Connection.

A reference to the Neuron at the start of this Connection.

Type:
Source:

target :Neuron

A reference to the Neuron at the end of this Connection.

A reference to the Neuron at the end of this Connection.

Type:
Source:

weight :number

The weight is used as a multiplier for two purposes.

The weight is used as a multiplier for two purposes. First, for activation, when transferring the output of the source Neuron to the input of the target Neuron. Second, during training, calculating the total error delta.

Type:
  • number
Source:

Methods

accumulate()

Calculate and accumulate gradient.

Calculate and accumulate gradient. Does not update weight.

Source:

update()

Update weight and reset accumulated gradient.

Update weight and reset accumulated gradient.

Source: