Is there any way to control the extraction of features?How to recognize what features are been learnt during training i.e relevant information is been learnt or not?