Supervised Learning (Machine Learning Lesson #2)

Stage 1: Say we play a game to get a baby to recognize images of cats. We have a soundboard right next to the baby: with two big round buttons. Button 1 and Button 2 light up into green and red respectively when pressed. We have several cards, each of which has an image of one animal.

Stage 2: We show the baby one card at a time, and if its a cat, we light up Button 1 which also makes a “Meow” sound (Success). If its another animal, Button 2 lights up and makes a “Boo” sound (Failure).

We repeat this process showing the baby several images (T).

Stage 3: Now we show the baby a new card but do not light up any button, or enable any sound from our end. We do not give any output. The baby recognizing that something is missing, presses the button that she thinks needs to light up. We say that the baby is learning if it is able to improve its chances to identify the right button with every new card correctly.

This is an example of supervised learning. Let us try to generalize this a little

Stage 1 is the setup, where we make sure the infrastructure (sourcing the data, loading it, processing it among others) is all set.

Stage 2 is the learning stage, where we use training data (T) to train our model (the baby). Interestingly, we know which card is a cat and which one is not, and are helping the baby (model) learn from those answers. Thus supervising the learning process (by #training providing the right answers for the model to learn from) and thus we call it supervised learning.

Stage 3: is when we use our trained model/machine to give us the output (O) for previously unseen data as an input.

There are different types of supervised learning models, where the output (O) value types are different. We will cover that in future posts.

Some examples of supervised learning: based on past information about spams filtering new incoming emails, based on prior knowledge prediction of weather