Recently, I started studying Machine Learning being from a programming background I choose fast.ai to be my headstart but after watching few videos i find fast.ai to be more of a tooling (high-level api ) knowledge.
Being not able to visualize why CNN, RNN, Batch Normalization, etc. are created. I just go by the blind approach of applying CNN to image problems and RRN to every sequential problems but i had found that for some case we can use RNN in images ( only relating image captioning ) and CNN for NLP.<p>To make my knowledge more concrete i started reading deeplearningbook ( by Ian Goodfellow ) and it is a great book for getting in-depth understanding but this book is an only theory (no practical code) and i seem to forget the concepts after a week or so.<p>So my question is how do you retain the concepts or tricks ( like initializing hyperparameter based on activation function used, etc. ) for practical projects?
Or what your approach to learning this overwhelming technology?