Oblivious Machine, Thou Shalt Learn

Why do we want machines to learn? What do we want to them to “learn” exactly? There is a lot of misunderstanding regarding the concept of machine learning. Is machine learning just robots learning how to walk and talk? Is it just machines learning how to do a particular fancy task? Not exactly. I thought I should just try to explain it in plain simple terms. Think about the following: How do search engines provide the best possible search results? How does multi-touch work on your smartphone or tablet? How can speech recognition systems understand us? How do social networking sites provide relevant suggestions? The answer to all these questions and many more is machine learning. Machine learning lies deep within our needs and affects our everyday lives, it’s just that people are unaware of it.

Wait a minute, I’ve seen this before!

A machine is like a newborn child. It knows nothing and it cannot do anything on its own. When we have large tasks at hand, doing them manually is not an option. We want machines to be able to learn on their own and make intelligent decisions based on the conditions. Let’s consider a baby named John. Just like how John looks around and learns from his surroundings, a machine also needs something to learn from. This comes in the form of input data to the machine. The data we feed to the machine is used by it to learn and understand the behavior of the data. Generally, the elders around John lay down a set of rules and tell him what’s right and what’s wrong. John can make a decision based on these rules when he is faced with a similar situation. To put it in machine learning lingo, this is called Supervised Learning. Somebody is telling you exactly what’s right and what’s wrong. Some of the popular applications include Face Recognition, Fingerprint recognition, Suggestions given by websites, Multi-touch gestures on gadgets and many more.

Where am I? What is this?

What if John is in a situation which he has never encountered before? There are no rules to follow because such a rule doesn’t exist. In this situation, John just has to look at the surroundings and make the best decision. Let’s say he wants to eat something. He looks around and sees apples, oranges and some other fruits in a bowl. John will infer that the fruits belong to one category, and the bowl probably doesn’t belong to this category because it looks so different from the other items. In this way, John learns and categorizes the items based on their inherent nature. In machine learning, we call it Unsupervised Learning. Nobody is teaching you how to make your decisions. Some of the popular applications include Medical Imaging, Genetic Clustering, Market Analysis, Search Engines and many more.

Supervised and Unsupervised Learning are the two extreme ends in the spectrum of machine learning. Is there something in between? Well, there are a few categories here but I will pick the one which is a good representative of them.

Fool me once, shame on you. Fool me twice, shame on me.

Consider the earlier situation. Now what if there is a red colored ball in the bowl along with other fruits? John will probably think that the ball is a fruit too. Let’s say John picks it up and tries to eat it. He will not be able to bite it because its too hard. John has never been in this situation before, so this should fall under the category of Unsupervised Learning. But does that mean John will try to to bite it every single time he comes across a ball? No. The next time he comes across a red ball, he will understand that it’s not a fruit and he should not eat it. Nobody is around to lay down the rules for him. He learnt it on his own based on his experience. He has to keep this in mind the next time he comes across a similar situation. In machine learning lingo, this is called Reinforcement Learning. Some of the popular applications include Autonomous machines exploring unknown terrains (like cars or robots), Telecommunication networks, Sensor Networks, Finance and many more.

Most of the machine learning algorithms can be classified into one of these three categories. There are a few others as well, but that will take another full blog post to explain. Learning machines make our lives a lot easier and machine learning helps them do their job well.

————————————————————————————————-