Machine Learning - an Introduction

What is Learning?

We start with the most basic of all questions when it comes to Machine Learning. What exactly is Learning? By the term Machine Learning itself, we might understand that it implies that some Machine is learning. What this machine could be is a question we’ll dive into later. But first and foremost, what exactly is Learning?.

Consider a Learner. This learner could be an inanimate entity (like a machine) or a person or any other animal. For our discussion, we’ll consider the learner in an abstract sense. This learner is also supposed to do a Task \(T\) and while doing the task \(T\), it is gaining Experience \(E\) on \(T\). We are also measuring its Performance \(P\) on the task \(T\).

DEFINITION

We say that a learner is learning to do a task \(T\) if its performance \(P\) on the task \(T\) increases as it gains experience \(E\) on the task \(T\)

In Machine Learning this learner is a Machine, usually a computer or more specifically a CPU/GPU. This is generally how machine learning is defined.

The Learner in Machine Learning

In most of the Machine Learning Algorithms that we’ll study later, the learner is essentially a machine that takes an input and gives out an output.

The task of this learner is to correctly match or give out an output when it is provided with an input. We generally calculate its performance based on how correctly it gives out a set of outputs given a set of inputs. The experience is provided in the form of Data (essentially a bunch of information) about the inputs and outputs by a Trainer.

Supervised v/s Unsupervised Learning

The learner that we described in the previous section can be trained in two different ways based on how we provide the data about the inputs and outputs.

When we provide Labeled data to the learner (i.e. data with set of inputs and their corresponding correct outputs), we say that the learner is learning in a Supervised manner. As the trainer is supervising the learning process of the learner, we call this Supervised Learning.

On the contrary, when we provide Unlabeled data to the learner (i.e. data with set of inputs but not their corresponding correct outputs), we say that the learner is learning in an Unsupervised manner. As the trainer is not supervising the learning process of the learner, we call this Unsupervised Learning.

NOTE

Here, supervision means that the trainer knows beforehand what outputs to expect for some given inputs.

Online v/s Offline Learning

The data provided to the learner by the trainer can either be in the form of small sets sent over a period of time or it can be sent all at once. The former is known as Online learning while the latter is called as Offline learning.

Concluding Remarks

That’s all for today’s blog post. We learnt about the basics of machine learning today and also how it is categorized as well as a basic layout for the learner. In the next blog post, we’ll learn more about supervised learning