elliotsig

(To be removed) Elliot symmetric sigmoid transfer function

elliotsig 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

A = elliotsig(N)


Description

Transfer functions convert a neural network layer’s net input into its net output.

A = elliotsig(N) takes an S-by-Q matrix of net input (column) vectors, N, and returns an S-by-Q matrix, A, of output vectors, where each element of N is squashed from the interval [-inf inf] to the interval [-1 1] with an “S-shaped” function.

The advantage of this transfer function over other sigmoids is that it is fast to calculate on simple computing hardware as it does not require any exponential or trigonometric functions. Its disadvantage is that it only flattens out for large inputs, so its effect is not as local as other sigmoid functions. This might result in more training iterations, or require more neurons to achieve the same accuracy.

Examples

Calculate a layer output from a single net input vector:

n = [0; 1; -0.5; 0.5];
a = elliotsig(n);

Plot the transfer function:

n = -5:0.01:5;
plot(n, elliotsig(n))
set(gca,'dataaspectratio',[1 1 1],'xgrid','on','ygrid','on')

For a network you have already defined, change the transfer function for layer i:

 net.layers{i}.transferFcn = 'elliotsig';

Version History

Introduced in R2012b

R2026a: elliotsig will be removed

elliotsig will be removed in a future release.

To learn more about using activation and transfer functions in neural networks, see these examples and topics:

• Build Networks with Deep Network Designer

• List of Deep Learning Layers

• List of Functions with dlarray Support

• Compare Activation Layers

• Example Deep Learning Network Architectures

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