Create feedforward ANN specifying number of input /output variables, hidden layer, weight and bias vales without dataset

We want to create feedforward net of given topology, e.g. one input layer with 3 nurone, one hidden layer 5 nurone, and output layer with 2 nurone. Additionally, We want to specify (not view or readonly) the weight and bias values, transfer functions of our choice. The feedforward net (without training) will be used to predict 2x1 output from 3x1 input. Basically Y=LWi*tansig(IWi*X'+bias1)+bias2 serves our purpose, but we want it is wrapped in a networkobject not as eqn. We donot want to configure it by presenting dataset in hand, which we donot have. Looking a guidence where we can use a syntax similar to this net = network(numInputs,numLayers,biasConnect,inputConnect,layerConnect,outputConnect)

Answer 1

[Sanjana]

Hi,

I understand that you need help with creating a custom feedforward neural network.You can use the “network” function in MATLAB, and create a feedforward ANN using the below syntax,

net=network(numInputs,numLayers,biasConnect,inputConnect,layerConnect,outputConnect)

Please refer to the below link for detailed explanation,

https://www.mathworks.com/help/deeplearning/ref/network.html

Hope this helps!

[Biswanath Mahanty]

Thank you Sanjana, I am already aware of this network function. But could not really figure out. A very small example e.g., 3 predictors, two hidden layer with 4 neuron, and two output will be of great help.