Deep Learning is the subset of Machine Learning.
People think that Deep Learning can do anything.
Then why do we need Machine Learning? When we use Machine Learning?
I have my opinions here,
- Deep Learning CAN'T do anything
Deep Learning models have lots of layers inside it. So they say that deep Learning can do anything with deep neural networks.
It is not true. The targets and sample points in deep learning should be the elements of manifold space. This means there should be a path from one point to another in sample space. For example, in manifold space, there is a path from number 1 to number 7 so that we can classify number 1 and 7.
So if the sample points are not in the same manifold space, the deep learning model does not work.
- Why do we use Machine Learning?
Deep Learning is not perfect and can't be the solution to all of the problems. My opinion is that we can use Machine Learning models when we can guess the relations between input and output. Humans can guess simple relations by themselves, but if the relation is complex, then it is impossible. We use neural networks, which have many layers that can describe the complex relations between input and output.
This is only my opinion, so I hope the readers give me your opinion and suggestion.
I am working with ML/DL, and I hope to grow with all of you, share knowledge, and speed up the improvement.
Thanks a lot.