Motivation :

Deep learning has shown an amazing performance on image/audio processing. However, for very deep model, vanishing/exploding gradient problem remains, because of learning method for deep model, back propagation.

For given dataset, which layers (low, middle or high) are needed to deep is different.

Thus, I have researched automatically finding the best fitted deep model for given dataset except vanishing/exploding gradient problem.

Performance :

I just tested this software to simple dataset (mnist, cifar10), and found the SW can find almost the best fit deep model for each dataset.