Machine Learning is a subset of Artificial intelligence. Artificial intelligence refers to machine intelligence and how machines can observe, learn, and improvise like humans. 

An intelligent machine is capable of observing from its surrounding environments (using sensors), learn from examples (past behavior), and making decisions without taking orders from humans or other devices.

Example: A vacuum cleaner who can identify nearby objects as debris or not by itself without any intervention by humans. 

Now, we know what artificial intelligence is. But, where does Machine learning fits in all of this?

Machine learning is a subset of Artificial intelligence. If AI is a person, then machine learning can be thought of as an organ of his body. We can take the example of a vacuum cleaner ( an artificial intelligent agent) who can classify objects as debris or not with the help of machine learning. 
Therefore, we can say that machine learning is artificial intelligence (a very important part) but artificial intelligence is not machine learning. 

Wait, what 😓 ?
Yes, Artificial intelligence is not just machine learning. Robotics, Rules engines, knowledge graphs, and expert systems can all be considered as AI.

Machine learning as stated by Tom Mitchell is a set of methods that learn by experience without being explicitly programmed. 

Formal Definition: 
A machine is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.
- Mitchell, Tom. Machine learning (1997)

Machine learning takes a method and data to learn from data. Data need not be of a particular type, ML algorithms can take multiple forms of data and learn from them. 
For example, Naive Bayes (ML algorithm ) can be used to classify an image (pixels) as a cat or a dog and can also be used to classify an email (text data) as spam or not spam.

Machine learning can be further classified into four major types: 
  
Supervised Learning:  

Learn with past examples (learning with the help of a teacher).  
Example: Consider you have a set of images (dog or cat), but you do not know for any particular image whether it is a dog or a cat. Now, how will you identify?
Hint: You can take help from a teacher (expert) but only once.

In the case of supervised learning, assume you have a teacher who is ready to help you with your task. This teacher will take a subset of your images and label them as dogs or cats. This set of images is called as training data. You can assume your teacher has a very good understanding of the matter and his labels are correct. 

Will you be able to identify the rest of the unlabelled images (unseen data/ testing data)? 

Now, you would take the labeled images and find a function mapping from inputs(image) to outputs (dog/cat). This is an example of supervised learning, learning with a teacher, or learning from past experiences (or labeled data ). 
Note: In supervised learning, the algorithm learns from labeled training data. 

Examples:
  1. Classification
  2. Regression 

Unsupervised learning: 

Learning without a teacher (no labeled data ). 
Example: Consider you have a set of images (cats and dogs only), but you do not know for any particular image whether it is a dog or a cat. Now, how will you identify?
Note: Teacher is on vacations and cannot help you 😅

In the case of unsupervised learning, you would take the images and find patterns in data that can help you cluster images into two categories (a dog or a cat ). Most of the human and animal learning is unsupervised, learn by observation.
Note: In unsupervised learning, the algorithm learns from unlabelled data. 

Examples:
  1. Clustering    

Semi-supervised learning: 

Example: you have a set of images some of which are labeled and rest are unlabelled. 
Semi-supervised learning is a midpoint between two extremes, supervised and unsupervised learning.
In semi-supervised learning, training data includes both labeled and unlabelled images. So, your algorithm learns from both labeled and unlabelled data.


Reinforcement learning:

Example: Playing a game. Initially, you do not know anything about the game. you take a few random guesses and receive a reward for each action (reward could be positive or negative ). Based on the rewards received, you will gradually learn which actions will improve your reward.

Examples:
  1. Robotics for industrial automation
  2. Multi-armed bandits