• “How do I end up recognizing a leaf?”, a pedestrian asked me yesterday.
  • Surprisingly, that is not a tedious question. I’m interested in hearing how have you exposed this subject to that ordinary someone, a complete stranger to this kind of matters.
  • I have started by saying that in order to learn the meaning of the word ‘leaf’, therefore recognizing one; is necessary to be presented in front of you a couple of particular leaves. That way, the idea of “leaf” will form in your mind.
  • Ok. So this idea you’ll end possessing is practically a general picture, you’ll see what is in common among the presented leaves.
  • Yes. If you’ll be asked, you’ll be able to say what are the properties or attributes that all this leaves have in common, like that they are of color green primarily.
  • So we are inclined to think that this general idea of a’ leaf’ is something like a visual image, but one which only contains what is common to all leaves.
  • What is important is that after this presentation, you’ll be easily capable to recognize any other actual leaf; only after you saw a couple of them.
  • That is fortunate, this ability to generalize; where the computer equipped with machine learning software needs to be presented with thousands of different images of that particular subject.
  • It seems that we are missing something. As you know and said, in order for a computer/machine learning software to recognize a cat in a photograph for example, it needs to see thousands of photographs with cats in different postures, of different colors, of different breeds and so…
  • But in the end is capable to recognize a new photograph of a cat, that it haven’t seen, still by some kind of generalization or interpolation.

Just expressing some whimsical ideas. Attention! The characters that will be presented are pure fiction and the conversations didn't take place.