*NEURAL NETWORK DATA

Define the bulk data for a neural network.

This option is used to define the internal parameters of a neural network. It can be used only as a suboption of the NEURAL NETWORK option.

This page discusses:

See Also
*NEURAL NETWORK
In Other Guides
About Neural Networks
Evaluating Abaqus Neural Networks

Products Abaqus/Standard

Type Model data

LevelModel

There are no parameters associated with this option.

 

Data lines to define the neural network weights and biases for a feed-forward, dense neural network (NEURAL NETWORK, TYPE=FFD)

First line
  1. Layer number.

Subsequent lines
  1. Weight values for the layer followed by the bias values (if the layer used biases). The number of values needed depends on the size of this layer and the size of the previous layer (or the network input size, if this layer is the first layer). Provide all weights for the first neuron in this layer with respect to all outputs from the previous layer, followed by all weights for the second neuron in this layer with respect to all outputs from the previous layer, and so on for each neuron in this layer. Then provide the bias values for all neurons in this layer (if the layer uses biases). You might need to use multiple data lines (with eight values per line) for providing all weights and biases for each layer. Weights and biases are assumed to be dimensionless.

Repeat this sequence of data lines as often as necessary to provide the layer number followed by the weights and biases for all layers in the neural networks. The number of values for each layer must be consistent with the configuration of the neural network defined under the parent NEURAL NETWORK option.