If you don’t happen to spend all of your free time obsessing over logistic regressions, gradient descents, and big data, but still want to know what machine learning is and why it’s such a big deal, then this guide is for you. Read on to find out what all the hype is about and why machine learning is one of the most exciting things to get into right now.

🤖 Table of Contents 🤖
What is machine learning?
Machine learning: the basics
Why machine learning matters
The future of machine learning
Further reading

What is machine learning?

Machine learning is a sub-field of artificial intelligence that uses algorithms to make computers learn from data, without being explicitly programmed to do so. These algorithms use statistics to find patterns in massive amounts of data and in doing so, the algorithms are able to change and improve themselves, eventually without human input.

Generally, this data takes the form of observations of real-world interactions, such as words, pictures, clicks, online transactions, or search queries. Basically, if it can be stored digitally then it can be fed into a machine learning algorithm.

For instance, we can teach a machine to accurately tell us whether the images below are of a) a cuddly feline or b) the thing you grab from the freezer after a tough day at work:

Indeed, people are training computers to tell the difference between labradoodles and fried chicken, chihuahuas and muffins, and pugs and loaves.

While differentiating between food and pets might seem trivial, the ability to correctly label an image without being explicitly programmed to do represents an important development in artificial intelligence.

Machine learning: the basics

Machine learning is based on three different approaches: supervised learning, unsupervised learning, and reinforcement learning.

Supervised learning

A supervised learning algorithm takes a known set of data (input) and known response (output) and then trains a model to generate a prediction on a new set of data. Supervised learning should be used when you have known data that you’re trying to predict.

Types of supervised learning

There are two main methods of supervised learning to be familiar with, regression and classification.

Regression is used to predict continuous responses that are measured along a sliding numeric scale. This could be changes in house prices, temperature fluctuations, or profit. A common regression algorithm is linear regression.

Classification methods predict discrete responses, where the outcome can be placed into a category or classification. For example, a classification method could be used to determine whether an email is or is not spam, whether it may or may not rain tomorrow, or whether a political candidate is or is not likely to win an election.

In contrast to a regression algorithm like linear regression, which looks at the relationship between variables on a sliding scale, classification works best when data can be categorized or separated into specific, distinct groups. Two popular classification algorithms are logistic regression and support vector machine (SVM).

Unsupervised learning

Unsupervised learning works by taking unlabeled data and finding intrinsic structures, or patterns, within that data.

Types of unsupervised learning

There are two primary methods of unsupervised learning, clustering and data compression.

Clustering is used to explore data and find hidden patterns. This could, for example, be used by companies to better understand their customers or by scientists trying to develop new drugs. A common algorithm for clustering is k-means clustering.

Data compression techniques perform something called dimensionality reduction which basically removes redundant and non-useful information from a dataset to minimize the amount of data that an algorithm has to process. Principle component analysis (PCA) is a really common data compression method.

Reinforcement learning

Reinforcement learning algorithms learn through trial and error. Like when training a dog, the algorithm is rewarded or penalized based on its performance towards reaching its objective.

Reinforcement learning is behind Google’s AlphaGo, as well as the recent advances in self-driving cars

Why machine learning matters

In case you weren’t convinced that interest in machine learning is currently skyrocketing, check out this chart that shows Google searches for the term “machine learning”:

Machines that learn are all the rage right now because they’re on the forefront of work into creating intelligent machines and are able to find patterns much faster and more reliably than humans can. In turn, they can help us solve more problems and make better-informed decisions on some pretty complex challenges.

As the amount of data that humans produce continues to grow at an ever-increasing rate (to the tune of 2.5 quintillion bytes of data every day), the ability to work quickly and accurately through massive quantities of data is becoming a necessity.

Because these algorithms are playing an ever-increasing role in our daily lives, it’s important that as many diverse voices are involved with machine learning and artificial intelligence as possible.

Since algorithms only do what they’re trained to do, these systems need to learn from as many different sets of data as possible, which relies on the discretion of the person selecting and feeding in the data.

With this in mind, having a wide spectrum of people involved in selecting, processing, and analyzing the data will help to make algorithms that can effectively respond to the infinite complexities, situations, and possibilities of real-world scenarios.

Otherwise, they will make potentially costly, offensive and even life-threatening mistakes.

The future of machine learning

Today, machine learning algorithms are used to:

  • Suggest songs to you on Spotify, shows on Netflix, and posts on Facebook
  • Help enable cars to drive themselves
  • Write and publish sport match reports
  • Identify faces in Facebook photos
  • Keep spam out of your inbox
  • Finish your sentence in Gmail
  • Recommend products on Amazon
  • Prevent financial fraud
  • Let you have conversations with Siri and Alexa
  • And the list goes on…

Basically, machine learning is all around us.

Despite significant advances though, there are some people who argue that the hype around machine learning is overrated and unwarranted. And to be fair, many of the algorithms aren’t even that new, they’re from the 70’s.

What has changed, however, is the power and speed of our modern day computers and the immense volumes of data we produce and have access to.

For example, in terms of processing power, your smart phone alone is more powerful than even the most powerful of supercomputers from 40 years ago.

At the same time, we’re collectively producing more data than in all of human history before us combined. And of course, machine learning is fueled by data.

The combination of more data and faster, more powerful computers has led to a surge in effort towards creating AI. But what does the future hold for machine learning?

Many are predicting more of the same, but bigger, faster, and cooler. This means more companies adopting machine learning techniques with more real world applications. This should also mean machines which are increasingly smarter and able to solve complex challenges on their own.

As the power of our computers increases so does the potential of machine learning algorithms.

At the same time, there is a push to improve interactions between humans and AI to the point where one day soon, we won’t even know we’re interacting with a bot. Google Duplex is nearly there.

While we’re still a ways away from reaching the singularity (if, indeed, we ever will), machine learning and AI continue to transform the way we work, consume information, and even the way we understand the world around us.

Further reading

Interested in learning more about machine learning? Here’s a list of some of the best guides and tutorials:

What is Artificial Intelligence?
What is Natural Language Processing?
Linear Regression: The Beginner’s Machine Learning Algorithm
DIY: Simple Linear Regression
Logistic Regression: Are You In or Out?