The easiest way to get started with machine learning

Do you use a personal assistant client like Siri or Alexa? Do you rely on spam filters to keep your email inbox clean? Have you subscribed to Netflix and rely on its amazingly accurate recommendations to discover new movies to watch? If you say "yes" to these questions, congratulations! You have made good use of machine learning!

Although this sounds complicated and requires a lot of technical background, machine learning is actually a fairly simple concept. To better understand it, let's look at what, who, when, where, how, and why about machine learning.

What is machine learning?

One day ladies will take their computers for walks in the park and tell each other, "My little computer said such a funny thing this morning".

—Alan Turing

At the heart of machine learning is "using algorithms to parse data, learn from it, and then make decisions or predictions about something in the world." This means that instead of explicitly writing a program to perform certain tasks, it is better to teach the computer how to develop an algorithm to accomplish the task. There are three main types of machine learning: supervised learning, unsupervised learning, and reinforcement learning, all of which have their specific strengths and weaknesses.

Supervised learning involves a set of tagged data. The computer can use a specific pattern to identify new samples of each marker type. The two main types of supervised learning are classification and regression. In classification, machines are trained to divide a group into specific classes. A simple example of a classification is a spam filter on an email account. The filter analyzes emails that you have previously marked as spam and compares them to new ones. If they match a certain percentage, these new messages will be marked as spam and sent to the appropriate folder. Emails that are not similar are classified as normal and sent to your mailbox.

The second type of supervised learning is regression. In regression, the machine uses previous (marked) data to predict the future. Weather applications are a good example of regression. Using historical data on weather events (ie average temperature, humidity, and precipitation), your mobile weather app can view the current weather and predict the weather in the future.

In unsupervised learning, data is unlabeled. These algorithms are especially useful because most real-world data has no tags. Unsupervised learning is divided into clustering and dimensionality reduction. Clustering is used to group based on attributes and behavior objects. This is different from classification because these groups are not what you provide. An example of clustering is to group a group into different subgroups (eg, based on age and marital status) and then apply them to targeted marketing programs. Dimension reduction reduces the variables of the data set by finding commonalities. Most big data visualizations use dimension reduction to identify trends and rules.

Finally, intensive learning uses the personal history and experience of the machine to make decisions. The classic application of reinforcement learning is to play games. Unlike supervised and unsupervised learning, reinforcement learning does not involve providing "correct" answers or outputs. Instead, it only focuses on performance. This reflects how humans learn from positive and negative outcomes. I quickly learned not to repeat this action. By the same token, a computer playing chess can learn not to move its king to the space that the opponent's pieces can enter. Then, the basic lesson of chess can be extended and inferred until the machine can beat (and eventually defeat) the top human players.

But wait, you might say. Are we talking about artificial intelligence? Machine learning is a branch of artificial intelligence. Artificial intelligence is dedicated to creating machines that perform complex tasks more than humans. These tasks typically involve judgment, strategy, and cognitive reasoning, which were originally thought to be the “forbidden zone” of the machine. Although this sounds simple, the range of these skills is very large—language processing, image recognition, planning, and more.

Machine learning uses specialized algorithms and programming methods to implement artificial intelligence. Without machine learning, the chess program we mentioned earlier would require millions of lines of code, including all edge cases, and all possible moves from the opponent. With machine learning, we can reduce the amount of code to a small fraction of the previous one. Great, right?

There is a missing part: deep learning and neural networks. We will discuss them in more detail later. Please note that deep learning is a subset of machine learning that focuses on the biology and processes that mimic the human brain.

Who developed machine learning? when and where?

A breakthrough in machine learning would be worth ten Microsofts.—Bill Gates

In my opinion, the earliest development of machine learning was the theory of the same name published by Thomas Bayes in 1783. The Bayes theorem found the possibility of giving events about historical data of similar events. This is the basis of the Bayesian branch of machine learning, which seeks to find the most likely events based on previous information. In other words, Bayes' theorem is just a mathematical method learned from experience and the basic idea of machine learning.

Centuries later, in 1950, computer scientist Alan Turing invented the so-called Turing test. The computer had to talk to one person through words, making people think she was talking to another person. Turing believes that only through this test can the machine be considered "smart." In 1952, Arthur Samuel created the first real machine learning program - a simple board game where computers can learn strategies from previous games and improve future performance. This was followed by the tic-tac-toe program that Donald Michie introduced in 1963. In the next few decades, advances in machine learning followed the same pattern—a technological breakthrough led to newer, more complex computers, often tested by playing strategic games with professional human players.

It reached its peak in 1997, when IBM chess computer Deep Blue defeated world champion Garry Kasparov in a chess game. Recently, Google developed AlphaGo, which focuses on the ancient Chinese chess game Go, which is widely regarded as the most difficult game in the world. Although Go was considered too complicated to be mastered by a computer, in 2016, AlphaGo finally won and defeated Lee Sedol in a five-game match.

The biggest breakthrough in machine learning is deep learning in 2006. Deep learning is a type of machine learning that is designed to mimic the thinking process of the human brain and is often used for image and speech recognition. The advent of deep learning has led to many techniques that we may use today (which may be taken for granted). Have you uploaded a photo to your Facebook account just to imply label the person in the photo? Facebook is using neural networks to identify faces in photos. Or Siri? When you ask your iPhone about today's baseball scores, your words will be analyzed using a sophisticated speech analysis algorithm. If there is no deep learning, it is impossible.