Examining how machines learn

- [Narrator] There are several ways machines learn, they're called techniques. Three common machine learning techniques are supervised learning, unsupervised learning, and reinforcement learning. Do you recall in the case study video when I mentioned the term supervised learning? Let's start there. Supervised learning is a technique that trains a machine learning model with labeled data. Labeled data includes known input and output data which teaches the machine to predict future outputs as new data comes in. Our case study to predict the cost of homes in California is a good example. The machine is trained on a data set that contains longitude, latitude, age, total rooms, total bedrooms, population, households, median income, ocean proximity, and the value we want the machine to learn how to predict the cost of the home. The machine studies the labeled data to find patterns and learns how to predict home values. This type of supervised learning is considered a regression problem because it is used to predict a number, also called a continuous value in mathematics. Another type of supervised learning we'll discuss solves classification problems. Classification techniques predict discrete responses like yes or no values. They are also great when you have data that needs to be tagged, categorized, or separated into specific groups or classes. Our case study to identify flowers is an example of multi-class label classification because we'll train the model to tag the type of flower found in an image. Another machine learning technique is unsupervised learning. Unsupervised learning doesn't use labeled data like supervised learning, instead it groups, or interprets data based only on input data. Clustering is a common unsupervised learning technique. The hospitality industry often clusters, or segments customers which allows them to tailor their marketing strategy to gain loyal customers and a competitive advantage. The last technique we'll discuss today is reinforcement learning which allows the machine to learn through trial and error. Unlike supervised and unsupervised learning, reinforcement learning does not rely on a defined data set, but operates in a dynamic environment that allows the machine to learn through experience. Reinforcement learning is similar to how you train a dog, with positive and negative reinforcement the dog learns the behavior that will get it the most treats or the most rewards. Robotics heavily relies on reinforcement learning. I'm sure you've heard of the AI programs that beat expert human players in board games, like Go and Chess while reinforcement learning is behind that. Machines learn in a variety of ways. Supervised learning teaches machines through labeled data while unsupervised learning groups and interprets data. Reinforcement learning allows machine to learn through trial and error. With all the ways machines learn they'll soon be smarter than us.