I'm trying to implement a sparse autoencoder with PyBrain. This is a simple neural network where the input and output are the same and you have one hidden layer with any number of nodes which is supposed to encode a compressed representation of the input/output. You enforce sparsity by introducing a penalty for the average activation of the hidden layer deviating from some small value, so that only a few hidden nodes are active at any time.
Is there any built in support in the trainers for this type of thing?
I'm trying to implement a sparse autoencoder with PyBrain. This is a simple neural network where the input and output are the same and you have one hidden layer with any number of nodes which is supposed to encode a compressed representation of the input/output. You enforce sparsity by introducing a penalty for the average activation of the hidden layer deviating from some small value, so that only a few hidden nodes are active at any time.
Is there any built in support in the trainers for this type of thing?