End to end neural network

When learning neural networks , I often see in the paper that , This network is end-to-end Of ,end-to-end Refers to what kind of network ？

End to end means that the input is raw data , The output is the final result , The original input is not the direct raw data , It's a feature extracted from the raw data , This is particularly prominent in the image problem , Because there are too many pixels in the image , High data dimension , There will be dimensional disasters , So the original idea is to manually extract some key features of the image , This is actually a dimensionality reduction process .
So here comes the question , How to mention features ？ The quality of feature extraction is extremely critical , Even more important than learning algorithms , for instance , Data classification of a series of people , The classification result is gender , If the feature you extract is the color of your hair , No matter how the classification algorithm , The classification effect will not be good , If the feature you extract is the length of hair , This feature will be much better , But there will still be mistakes , If you extract a super feature , For example, chromosome data , Then your classification will not be wrong . That means , Features need enough experience to design , This is becoming more and more difficult with the increasing amount of data . So it came out End to end networks , Characteristics can be learned by yourself , So feature extraction is integrated into the algorithm , No need for human intervention .
end-to-end The essence of is that the problem you want to solve is multi-stage or multi-step ( With the so-called raw feature Do not have what relation ). If you study in stages , The optimal solution in the first stage can not guarantee the optimization of the problem in the second stage .end-to-end Put them together to optimize , Ensure that the solution in the final stage is optimal .
The advantage is convenience .
The disadvantage is that it is more and more like a black box .