How does a machine learn

Learning starts from mimicking, as funny as it sounds, it is till a certain extend true. Humans observed the flight of birds and made an aircraft, then by observing the whales we made the submarines which are shaped like whales and submarines also use their sonar system to trace food and avoid hurdles. This is how we humans learn from mimicking. Advance machine learning is trying to do the same thing, making machines that mimic the human mind, specifically the decision making pattern.

Learning never stops, said someone wise. So why only limit learning to the human mind. Mimicking our ways of learning, machine too can benefit from it and if the machine benefits from it, ultimately we human benefit from it.

Most of the times we assume learning depends on “what” we learn, how hard or easy it is to grasp and retain. But recent study, shows that its about “how” we learn that determines if the information is easy or hard. Whether we have an idea about the data, that we want to predict or whether we are impromptu about the prediction. Here are the two types of learning for a model : supervised learning and unsupervised learning.

First lets discuss supervised learning. When we give input and output to the model and the model identifies the patterns, learns the trends and the association of the features, this is called supervised machine learning technique. Now when we give this trained model, new input data which was not used before, it will use all the information it has learned from the previous given input and output series of data and then it will return with the appropriate output. Example are “linear regression”, “decision tree classification”. Real life example: Mom asked you to get some groceries, and you are a supervised learner so you will, get yourself familiar with how the vegetables, fruits, grains look like and also note down their current prices. Now when you go out, you will better identify the right product and price.

Unsupervised is the exact opposite of the above. Here we give the data to the model and it directly gives the output. No previous input and output series of data is given here. So you may wonder how does the model learn, well here is a simple analogy. If you are going out for grocery shopping for the first time and you have no idea about the rates of vegetables, the amount which is required and so on. But if you continue grocery shopping for a couple of more times, you will start to understand certain patterns, like which vegetable is pricey, which fruit is to be taken in which season, how much do you need red chilies and most importantly which products are free. So here you learn from experiences, lessons, failure and so on.

Similarly when a model is given certain input it will generate its output and over the course of time, the model will learn from its mistakes, lessons, patterns, risks, association, etc. This is unsupervised machine learning technique. Example are : “k means” clustering algorithm and “apriori” association algorithm.

We also have the reinforcement type of machine learning, which I will discuss in the next article.