What is Machine Learning?

Machine learning is a branch of artificial intelligence (AI) and deals with how to make computers learn (model of relationships, reasoning and analogies) from experience so that computers can act in a smarter way without being explicitly programmed. Machine learning is an interdisciplinary field which borrows techniques and methods from computer science, statistics and engineering.

People often confuse machine learning with AI and data mining. Therefore, it is important to understand the difference between AI, machine learning and data mining. AI is a broader concept which deals with building of intelligent agents. Such agents are assumed to act like humans, e.g. self-driving car, robot chef, personal assistant etc. AI applications are often enabled by machine learning models trained on historical data about a problem. In data mining, we often work with techniques inspired from machine learning and statistics. These techniques help analysts to figure out preliminary insights and know what a massively complex dataset is about. However, data mining does not deal with uncovering of underlying trends and patterns, and building of models on top of those patterns. This is where machine learning kicks in.

Machine learning typically deals with three kinds of learnings:

1. Supervised Learning

In supervised learning, we learn a model (mapping function) based on a set of training examples where we already know actual answers for those examples. For example, imagine a real estate business which has 25-years historical data of house sale price. That means for each training example, we know know input attributes (area, number of bedrooms, number of washrooms etc.) and output (price at which it was sold). Therefore, this problem qualifies for supervised learning in order to learn a house prediction model.

2. Unsupervised Learning

We often use unsupervised learning when we do not have actual answers (output attribute) in training data. Imagine customer dataset of an online store, where we do not have any output attribute but we still may be interested to learn structure of customer data i.e. new customers, loyal customers, high spending customers etc.

3. Reinforcement Learning (RL)

Reinforcement learning is also know as goal oriented interactive learning. There is no right or wrong answer but a reward for every action. That reward helps learner to discover appropriate actions or behaviour. An intuitive RL example would be training of your pet; when pet listens to your actions you would give him/her a candy. At the same time, if pet messes with your living room or kitchen, you would hold back candy (reward). This would allow pet to learn good behaviour.