What is Machine Learning?

"Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to effectively perform a specific task without using explicit instructions, relying on models and inference instead."

Definition of Machine Learning according to Wikipedia

I'd prefer to define it as the technique of making an algorithm learn structures in data without explicitly programming it to do so. This is kind of what the above definition states.

Machine Learning is a subset of a larger domain known as Artificial Intelligence, in which you try to make an intelligent system. By that, I mean to make a system that can perceive an environment full of variables and can take actions that maximizes the chance of success. These skills in particular are in high demand, because they are extremely important in various fields such as:

• Automation: We don't just want automation, we want intelligent automation. One which can take decisions in unforeseen environments and deliver the best results. Robotics is increasingly becoming an important field day by day.
• Data Science: We generate enormously unimaginable amounts of data, every day (source). Big Data analysis is a set of methods where we find patterns and correlation between different variables in data. This skill is widely used in various more fields such as advertisement and marketing, research and development, medical industry, etc.
• Internet of Things (IoT)

The possibilities are endless...

In machine learning we develop algorithms and statistical models to perform tasks, basically generate some output. We essentially build a mathematical model for intelligence. That, in a very small essence, is what machine learning is. Using mathematical models to learn from data.

Machine Learning is sub divided, based on various different cases. There are three major sub groups into which all applications fall into, let's have a one by one overview of them

Supervised Learning

In this, we use a labelled dataset and we have to develop a model to gain insights from that data so that, if it's given variables that it hasn't seen in the dataset, it must be able to predict the output to a reasonable degree of accuracy.

A dataset here is basically a set of input and output pairs over which an algorithm learns to predict the output given an input. Examples include but are not limited to:

• Spam detection
• Optical Character Recognition and various fields of Computer Vision
• Housing Price predictions
• Handwriting, Speech and Pattern recognition
• Detection of tumors in the medical field
• Suggestion algorithms like the ones which Netflix uses to predict if a user would like a movie or what YouTube uses.

This is probably the most well known and used form of machine learning, given the amount of data that we have and continuously generate.

Unsupervised Learning

In this, we have a dataset, but not labeled. That means we have a bunch of inputs but no clear decisive output. We basically have to draw inferences from the dataset. The main purpose is to find groups or hidden patterns in data.

A dataset here is basically a set of input values, over which an algorithm must find correlation and grouping between various fields of data. Examples include but are not limited to:

• Clustering used widely to form customer groups, a nice example of marketing and data science
• Anomaly detection

This is another widely used algorithms for finding relations in datasets.

Reinforcement Learning

In this, we develop an algorithm that can take sequential decisions to maximize it's gains. Basically, the algorithm (agent, formally) performs some actions in an environment and an interpreter rewards or penalizes the agent based on the outcome on the actions in the environment. This is usually more complicated than the ones mentioned above, but a few applications include:

• An algorithm playing chess game or many other archaic games
• Control theory
• Genetic algorithms

Wherever a sequential decision model is involved