I want to write denoised autoencoder with Java. My auto encoder was written as this.
This is implemented as simple neural network which uses backpropagation algorithm for optimisation, updating 1st layer weight matrix and 2nd layer matrix which is the same of the transpose of 1st layer matrix.
However it cannot obtain sufficient good accuracy. Even running gradient descent further, I cannot obtain better accuracy. So in order to find good implementations, I searched any other codes that implements denoised autoencoder. But I can find only the one which receives binary input vector even though I want to pass the continuous input vector to my denoised autoencoder.
By the way denoised autoencoder should be run with binary input vector such as [0,1,0,0,1] not [0.2,0.3,0.0,0.5]. Is there no way to pass the continuous input vector to denoised autencoder?
Please let me know the implementation of denoised autoencoder which can train reconstruct continuous input vectors.