What is Machine Learning? Well, Machine Learning is a technique which allows the machine to automatically learn and improve from experience without being programmed explicitly. So, you feed the data to the machine and it learns from the data and builds the logic based on that data.
Let’s take a real-life example of machine learning. Do you know how does Netflix recommends movies to its users? Well, obviously the answer is Machine Learning. But how? Netflix takes its user’s past behavior(user’s history) to learn what types of movies does the user likes and then recommends the movies based on that (I will not go into detail here).
So, why suddenly machine learning is gaining popularity? The algorithms we use in machine learning have been there for many years. The reason is that earlier neither we had enough data nor enough computational power to run these algorithms. Did you know that 90 percent of the data in the world was generated over the last 2 years alone? So, how do we make sense of this amount of data? Human mind is not capable to handle this amount of data.This is where machine learning algorithms are useful.
Machine Learning can be categorized into three categories depending on the nature of the learning:
In supervised learning, we teach or train the machine using labelled data i.e. data in which both the input and the desired output are provided. The machine then learns the function to map input data to the output data.
For example, let’s say that we have house prices data.
The input variables are:
- Area of the house
- Number of rooms in the house
- Age (How old is the house)
The output variable is:
- Price of the house
So, when we train our supervised learning model on this data,it learns a mapping between the input variables and the output variable, like the houses with more area will be priced higher and for old houses the price will be comparatively less. Now, if we give our model the input of any new house, it will be able to predict the price of the house.
Supervised Learning can be further divided into 2 parts:
Regression: A regression problem is where the output variable is a real value, such as “house price” or “height”.
Classification: A classification problem in where the output variable is a category, such as“cat” or “dog” or “healthy” or “not healthy”.
In unsupervised learning, we train the machine using unlabeled data i.e. there is only input data and no output data. The machine itself discovers interesting structures based on the input it receives.
For example, you have a set of news articles, unsupervised learning can group them into set of articles about the same story. So that, one group may contain articles about sports, other group may contain articles on politics.
In Reinforcement Learning an agent learns how to behave in an environment by performing actions and learning from its own experiences. Reinforcement Learning is growing rapidly and possesses immense potential.
For example, Let’s consider a board game like Chess. In order to play the game, the player need to think about multiple factors. If we were to build a machine to play this game, then to cover all the possibilities, we need to specify a large number of rules. In reinforcement learning we need not manually specify rules. The agent learns by playing the game. The agent plays the game multiple times and each time learns from its mistakes to get better at the game.
The applications of machine learning are rapidly gaining popularity and it is important to keep in mind that the algorithms and approaches will continue to change.