What is the difference between applying shallow-learning methods repeatedly and deep learning?

Question

In the book Deep Learning with Python, François Chollet writes (section 1.2.6, page 18)

In practice, there are fast-diminishing returns to successive applications of shallow-learning methods, because the optimal first representation layer in a three-layer model isn't the optimal first layer in a one-layer or two-layer model. What is transformative about deep learning is that it allows a model to learn all layers of representation jointly, at the same time, rather than in succession (greedily, as it's called).

By shallow learning, we mean traditional machine learning models that aren't deep learning, such as support vector machines.

I understood the above as below.

Using a model with three-layer shallow-learning methods has the same output (predicted) value as using one-layer shallow learning method. The effect of using multiple layers of shallow learning methods is to 'increase running time or repetition'.

Did I understand properly?

Answer 1

Quite surprising to find someone reading this same book. I read this part a week ago and the explanation is quite clear in the book :

• If you use successive shallow learning methods, you first train one model, then you train another model with the outputs of your first model, and then a third with the outputs of your 2nd model. The problem with that is that each model is trained to get good results at his task, not to send information, so there can be an increase when adding successive models, but it is a very weak increase.
• If you use deep learning, all the layers are trained at the same time, so each layer learns how to efficiently transfer important information to the next layer of the model. This is why it is much more efficient.

Hope I made it clearer

Answer 2

No, you did not correctly understood the meaning of the passage.

Using a model with three-layer shallow-learning methods has the same output (predicted) value as using one-layer shallow learning method, The effect of using multiple layers of shallow learning methods is to 'increase running time or repetition'.

A three-layer shallow learning method (so three one-layer methods stacked one after the other in a secuential way) has not the same predicted output value as a one-layer shallow learning method. It will be different output (and a 3 layers model should show some improvement) and clearly will need more time to get the result.

Deep learning methods as mentioned by @Ubikuity are (generally speaking) mode efficient at finding the important features and (usually) get better results than using a model with three-layer shallow-learning methods.