Definition Of Deep Learning In The Textbook

In the textbook, Deep Learning is defined as follows in Chapter 1:

Deep learning is a computer technique to extract and transform data.

I am trying to understand the two key words mentioned, extract and transform. What does it really mean to extract and transform data? How was this done in “Is It A Bird?” model.

Someone please shed some light on this.

1 Like
  • Feature Extraction
    • To be able to distinguish between images, there must be features that the model identifies. No one programmed the model what those features are, so the model must have “extracted” those by itself.
  • Transformation is…
    • You start with an image and end up with a label.

How was this done in “Is It A Bird?” model.

That would be the purpose of the whole course. You are at chapter one.
For the moment, treat the model as a blackbox. (For some that leaves a tension - in which case its a useful skill to practice.)

1 Like

Hi Samar

That is an unusual definition.
Extract Transform and Load ETL is the process of taking raw data and additing to a dataset so it can be used.

Deep Learning take data and creates a series of internal representations (latents) of the data and outputs an external representation.

So (house price, zipcode, near by school results, average resident income, future infrastructure plans, state of economy) → L1 L2 L3 → sales price. The idea being that the input data has hidden properties which can be determined to determine an output. We assume the input existings in n dimensional space but we can approximate it in m dimension space and so there exist a magic box that takes the m dimensions to produce the output. So for example I never included bad neighbours or near flood zone in my list but in some circumstance they could affect the sales price.

I would Jeremy’s book. It is very good.

Regards Conwyn

1 Like