Machine learning

Machine learning (ML) algorithms typically fall into three categories: supervised learning, unsupervised learning, and reinforcement learning.

In supervised learning, ML algorithms are used to classify existing data into some predefined categories, or to predict what categories new data might have. The algorithm is taught the categories in advance through labeled data, where each data point is given a label with its correct category. By training on the labelled data, the algorithm can learn the underlying patterns and predict what data belongs in which category.

In unsupervised learning, the ML algorithm is given unlabeled data. In its training phase, the algorithm learns what patterns exist naturally in the data. Typically, it will cluster similar data points together as per some predefined function of similarity. Unlike in supervised learning, the clusters found through unsupervised learning are not automatically associated with existing labels. A programmer or end-user of an unsupervised ML algorithm may need to further analyse the found clusters to further interpret them.

In reinforcement learning (RL), the ML algorithm will not have labelled data. Unlike in unsupervised learning, however, in reinforcement learning, the algorithm is given some type of feedback on the correctness of its output. This feedback is not necessarily given to the algorithm by a person, such as is the case with supervised learning (where the labels for the data are usually provided by human experts). Often, RL algorithm will have a feedback function built-in to the algorithm. During its training phase, the RL algorithm learns to decrease the errors it makes by continuously generating output, processing the correctness of that output through its feedback function, and adjusting how it works to ultimately increase its correctness.

Deep learning

Beyond machine learning, the term 'deep learning' is also often used within the field of Artificial Intelligence. Deep learning is form of machine learning that tries to overcome some limitations of machine learning. Deep learning (DL) algorithms perform unsupervised learning. By using large amounts of data, DL algorithms are capable of themselves learning the labels that supervised ML must be provided with. A field in which deep learning algorithms are often used is that of image classification, where images are autonomously classified as e.g. cats or dogs.