ELI5: What is Machine Learning?

Machine Learning (ML) refers to a branch of Artificial Intelligence (AI) that processes data to achieve a specified objective, and improves automatically through experience over time.

What is an algorithm?

An algorithm is a set of rules that a computer program can follow. For example, machine learning algorithms can be used to predict stock prices, act as your assistant or workout coach, help doctors make more informed decisions during diagnosis, or dispatch self-driving cars.

In general, a user of a machine learning algorithm specifies a performance measure (or metric) that is used to quantify how well it’s doing at its job. Subsequently, the task of the algorithm is to learn how to do better at this job by optimizing the performance measure. An example of a task for an algorithm may be to assign a data classification. If you use email you may notice that some emails are classified as spam and put in a separate inbox – this is a simple example of classification. Another common example is regression, which allows an algorithm to predict outcomes based on the value of one or more predictor variables, such as weather forecasts, which use different variables to predict what the weather will be like in the future. The next important element of machine learning is how the computer can learn.

How does machine learning work?

Machine learning uses computer algorithms, which simply put, are a set of steps for taking an input and producing an output. In machine learning these algorithms are typically used for either classifying data or making some type of prediction. To make these algorithms accurate the machine learning algorithm needs to be “taught” how to assess inputs to create good outputs. In general, there are three main categories for how this is done: supervised learning, unsupervised learning, and reinforcement learning. These are all different methods for training the computer algorithms to make good decisions and produce good outputs.

https://www.youtube.com/watch?v=f_uwKZIAeM0&ab_channel=OxfordSparks

What is supervised learning?

This type of machine learning is defined by the use of labeled datasets to train machine learning algorithms to classify data or predict outcomes accurately. In supervised learning, the algorithm is given a small training dataset to work with to give it a basic idea of the problem, solution and data points. The algorithm will develop the ability to find relationships between the given datasets and by the end, the algorithm will have an idea of how the data works and the relationship between inputs and outputs. Even after being deployed, the algorithm continues to improve on its ability to identify patterns. A common example of supervised learning would be your email service’s ability to classify spam into a separate folder in your inbox based on characteristics found in the email itself.

What is unsupervised learning?

Unsupervised learning uses unlabeled data to teach the algorithm to discover patterns that would be helpful for clustering or association problems. Once the algorithm has been appropriately trained it should be able to discover hidden patterns or data groupings without the need for human intervention. Since unsupervised learning algorithms do not need human intervention (through labels) they are better at adapting to different datasets. A real-world example of this is Google News, which uses unsupervised learning to categorize articles on the same story from different news outlets.

What is reinforcement learning?

This type of learning takes inspiration from the psychological concept of condition that humans use to learn. It allows for an algorithm to improve upon itself and learn in new situations using a trial-and-error method. The trademark of this type of learning is an interpreter and a reward system. As the algorithm is fed input and produces an output the interpreter will decide whether the outcome is favorable or not. Whenever a program finds a favorable solution the interpreter will provide it a reward (reinforcing the good behavior) while a non-favorable solution will force the algorithm to reiterate until it finds a better result.

Machine learning allows for computers to learn how to think and make decisions and predictions at a high degree of accuracy, even without human intervention. Through the advancement of training methods such as supervised learning, unsupervised learning, and reinforced learning, we have machine learning algorithms that can be used effectively in many areas of our lives. It’s because of these advancements that we have weather forecasting, email spam filters, targeted advertisements, and many more conveniences of the 21st Century.