initcon

(To be removed) Conscience bias initialization function

initcon will be removed in a future release. For more information, see Transition Legacy Neural Network Code to dlnetwork Workflows.

For advice on updating your code, see Version History.

Syntax

initcon (S,PR)


Description

initcon is a bias initialization function that initializes biases for learning with the learncon learning function.

initcon (S,PR) takes two arguments,

S	Number of rows (neurons)
PR	R-by-2 matrix of R = [Pmin Pmax] (default = [1 1])

and returns an S-by-1 bias vector.

Note that for biases, R is always 1. initcon could also be used to initialize weights, but it is not recommended for that purpose.

Examples

Here initial bias values are calculated for a five-neuron layer.

b = initcon(5)

Network Use

You can create a standard network that uses initcon to initialize weights by calling competlayer.

To prepare the bias of layer i of a custom network to initialize with initcon,

1. Set net.initFcn to 'initlay'. (net.initParam automatically becomes initlay’s default parameters.)

2. Set net.layers{i}.initFcn to 'initwb'.

3. Set net.biases{i}.initFcn to 'initcon'.

To initialize the network, call init.

Algorithms

learncon updates biases so that each bias value b(i) is a function of the average output c(i) of the neuron i associated with the bias.

initcon gets initial bias values by assuming that each neuron has responded to equal numbers of vectors in the past.

Version History

Introduced before R2006a

R2026a: initcon will be removed

initcon will be removed in a future release.

To learn more about initializing and customizing neural network learnable parameters, see these examples and topics:

• Compare Layer Weight Initializers

• Specify Custom Weight Initialization Function

• Neural Network Weight Tying

• Initialize Learnable Parameters for Model Function

For more information, see Transition Legacy Neural Network Code to dlnetwork Workflows.