[Pedagogy] Neural network pedagogy

It is really easy to get lost with neural networks. You mainly get lost in training process. You try to train the neural network with the data and you feel lost on how to use the network.

Through training the network, we are creating a blackbox function

By using the trained network, we are loading that trained model on to our system and use it as as function 

I found the following methods of making a note of neural networks to work exceptionally well.When given a neural network-related paper. You need to make sure you at least understand these 5 things below, with the mindset that we are making a black box (function estimator) that gives an output when you insert input. With that in mind, we need to check and make notes of these 5 things.

1. Is it supervised? non-supervised? (I reject the way how people define encoder/decoder as semi-supervised, it is still supervised)

2. What are its input and output?

3. What are you optimizing for in the model ?

4. What is the loss function?

5. After the training is done, what can the trained model can do ?

There should be more to a single paper than these 5. But just having these 5 written down while reading a paper on neural network helps me understand it better. If you have any other pedagogical thoughts to add on to this, please feel free to comment or send me an email.