Machine Learning, Deep Learning and Big Data is all the rage today. From more than 30 years of IT experience, it is clear that there are very little new in the world of compouters. Most of what we see are improvements on previous innovation, or using older innovations in new ways.

The same can be said for machine learning. It is the perfect confluance of powrful computers, massive amounts of data and some pretty old mathematical and statistical theory.

You can be excused for thinking this is revolutionary new technology, if you just pay attention to the hype in the industry and press. You will think that machines can actually learn, versus being taught.

Lets look at the way a baby learns. Say you take a drive out in the countryside, and every time you pass a horse, you point to it and go “look baby, a horse”. After a while, the baby will start to go “horse” every time you pass one. How did this learning happen? You had a bunch of data, and you labeled it for the baby. The baby learned what a horse looked like (4 legs, 1 head, 2 ears, tail, white/black/brown, rather large, eating grass in the field). It is likely to recognize a pony, or even a donkey as a horse. That is until you canm label enough of them for him, so that he can make the distinction himself.

And even though the baby learned what a horse loked like, it cannot recognize an elephant, or a plane, or a light pole until you have fed him enough samples with labels.

This page has the following sub pages.