We all get that machine learning is powerful and wonderful and magical (in spite of never having received an invitation to Hogwarts) but what problems can we solve using it? In a previous post, we discussed that AI was a blanket term describing intelligent systems (the entire house) and machine learning was the kitchen (where the really interesting stuff happens). We’d like to teach you all how to cook…

There are three basic sauces in the ML kitchen:

  • Supervised learning

  • Unsupervised learning

  • Reinforcement learning

Now, remember when your teacher stood over you, waiting for you to do some arithmetic? Similarly, supervised learning attempts to train an input space (the student) to achieve a particular output space accurately (“carrying the one” appropriately). Given a set of inputs, a model could develop and map these to outputs. These maps can create numerical correlations (guess how much I could sell my car for) or class correlations (guess my dog’s breed)

Unsupervised learning is what happens when the teacher steps out of class for a second – with no given outputs (authorities), the inputs (children) try to eat each other. Even in the midst of all these uncontrolled children, they tend to form clusters (cliques) and might become densely packed around a food source or the class pet. These insights prove to be quite valuable in the real world when trying to determine which emails are spam and the probability of there not being enough potatoes next season.

Santa and his questionable relationship with children is a prime example of reinforcement learning – be good and you get toys or be naughty and you get coal. Machines can be taught in a similar way (they are penalised for making incorrect decisions and rewarded for correct decisions). This way, a machine can learn (with remarkable accuracy) based on outcomes. This makes performance based learning suited to complicated scenarios in which all the inputs and outputs cannot be predetermined – like gaming or rain forecasting. 

These three sauces (we might’ve binged a little Babish before writing this) are phenomenal on their own. However, they can also be combined to create some interesting hybrids.

Follow your curiosity,

The Afrobots @ Afrobotix.

Afrobotix™ 2021