Artificial intelligence is an umbrella term that encompasses a wide variety of subfields that share the goal of enabling machines to perform tasks that require the use of human intelligence. One of these subfields is machine learning.
Machine learning is the science that allows computers to be programmed so that they can learn from data. In other words, we give a computer a certain amount of examples and it learns to recognize patterns and make predictions on its own.
There are many types of systems that help classify machine learning into broad categories, and one of them is based on the criterion of whether or not they are trained with human supervision. They are divided into supervised learning, unsupervised learning, and reinforcement learning. But what does each of these types of machine learning mean?
Supervised machine learning
Supervised learning is when the training set we provide to the model includes the labels that describe the class or category to which each example belongs. This means that each input example in our database is associated with a desired output. To illustrate, let’s mention Harry Potter and the Philosopher’s Stone.
During the movie, the teachers teach the students specific spells and magic. Drawing a parallel with supervised learning, the skills that the teachers teach the students are like the labeled data. The students act as a model in which they learn to map inputs to the correct outputs by casting spells correctly. Furthermore, they receive feedback on their spellcasting performance, which is used to adjust the model and improve its performance, or in this case, the execution of spells.
Ron and Hermione practicing the Leviosa spell
Ron e Hermione treinando o feitiço Leviosa
Unsupervised machine learning
On the other hand, in unsupervised learning, the training data is not labeled, and the system tries to learn without a teacher. The goal is for the algorithm to identify patterns and structures in the data on its own.
Going to the world of series, we can use the episode “Be Right Back” from the series Black Mirror as an example, in which the main character, Martha, uses an artificial intelligence program to create a digital version of her deceased boyfriend, Ash.
In this context, the program received text messages, emails, and various other online information available without any classification, and from analyzing this data, it learned enough patterns to replicate Martha’s boyfriend’s behavior.
Martha interacts with a synthetic recreation of her late boyfriend Ash
Martha interage com uma recriação sintética de seu falecido namorado Ash
Reinforcement machine learning
Finally, reinforcement learning is a very different technique in which the learning system has the ability to observe the environment, select and execute actions, and obtain rewards or penalties in exchange. It must learn on its own what the best strategy is to obtain the highest number of rewards over time.
A good real-world example is DeepMind’s AlphaGo program, which in May 2017 defeated world champion Ke Jie in the game of Go in the first match of a best-of-three. It learned its winning policy by analyzing millions of games and then playing many games against itself.
Which of these techniques should I use? It depends! Each type of learning has its own applications and the choice will depend on the specific needs and objectives of each project. Here at Hop AI, for example, we are able and are experts in working with all approaches to meet the needs of our customers. Want to know more about how we help our customers? Access our page about our offer on how to scale AI within organizations (Our Services) and get in touch if you are interested in starting this journey!