What makes a cat a cat?

If you don't know the answer, ask to a 3-year-old child, they can help you, but then try to 'program' it

From an input image of a cat, what vector of features can you get that could help you in the training process of a model?

• Image segmentation, border detection

• Points of interest, gradients

• Basic shapes detection

• Fast Fourier transformation and work in frequency domain

• etc.

What makes a cat a cat? From an array of pixels what can we do?

This has been the work of thousands of researchers for many years, being able to find some clever, and sometime very specific, techniques to get some characteristics from images (or other inputs) or features in order to train ML models.

Let's try to remove the human from the feature engineering process

With deep learning, the objective is to train from 'raw' data, and let the neural network to decide what feature to pay attention to