where are the initial weights and biases when training autoencoder?,

For example the autoencoder digits example which came with matlab

[xTrainImages, tTrain] = digittrain_dataset;

clf

for i = 1:20

      subplot(4,5,i);
      imshow(xTrainImages{i});
  end

% Get the number of pixels in each image

imageWidth = 28;

imageHeight = 28;

inputSize = imageWidth*imageHeight;

% Turn the training images into vectors and put them in a matrix

xTrain = zeros(inputSize, numel(xTrainImages));

for i = 1:numel(xTrainImages)

xTrain(:,i) = xTrainImages{i}(:);

end

hiddenSize1 = 100;

% Create the network. You can experiment by changing the number of training
% epochs, and the training function

autoenc1 = feedforwardnet(hiddenSize1);

autoenc1.trainFcn = 'trainscg';

autoenc1.trainParam.epochs = 400;

autoenc1.inputs{1}.processFcns = {};

autoenc1.outputs{2}.processFcns = {};

autoenc1.layers{1}.transferFcn = 'logsig';

autoenc1.layers{2}.transferFcn = 'logsig';

autoenc1.divideFcn = 'dividetrain';

autoenc1.performFcn = 'msesparse';

autoenc1.performParam.L2WeightRegularization = 0.004;

autoenc1.performParam.sparsityRegularization = 4;

autoenc1.performParam.sparsity = 0.15;

% Train the autoencoder

autoenc1 = train(autoenc1,xTrain,xTrain);

Where is the initial weights and biases of the autoencoder before training?

Answer 1

[Greg Heath]

Typically, train checks to see if weights exist. If not, then it will initialize the net.

Hope this helps.

Thank you for formally accepting my answer

Greg