Areas of Machine Learning and their real-world application
In a previous post we have learned that you will hardly find a successful app without an application of AI in the future, and Machine Learning plays a major role in that future. Machine Learning is the area of AI, where algorithms are not explicitly programmed for a task, but learn and improve by experience (data). There are many types of machine learning algorithms, some more known than others, that can be applied for different problems. Here we give a brief overview of some important fields and their applications.
Supervised Learning
In this setting we are given a set of labeled training instances and we would like to learn a model capturing the underlying characteristics within the data, s.t. we are able to make predictions about unseen instances. There are two main types, which are distinguished by the type of the labels.
The labels can be categorical, for example when we want to assess whether an incoming email is legitimate or just spam. Here the data we learn from consists of old emails for which we know the labels and we assign incoming emails into one of two classes, thus this is called a classification task.
If on the other hand the labels are continuous it is called regression. For example, given the price of the houses in our neighborhood, with some additional features such as the square footage and construction year, we can find the fair market value of our own house.
Unsupervised Learning
In unsupervised learning we have, as in the case of supervised learning, a data-set we learn from. The data, however, is not labeled. This task seems to be much harder because the machine is not given the label information present in the supervised setting.
Clustering is the most prominent example of unsupervised learning. Here we seek to discover the inherent grouping within the data. For example, we would like to group our customers to predict purchase behavior or churn probabilities, or to find the right approach to give incentives based on the characteristics of the group the customer belongs to.
Reinforcement Learning
In Reinforcement Learning, in a nutshell, an intelligent agent, e.g. a mouse in a maze, tries to find the right series of actions (move left/right/up/down) to reach its goal the cheese. To do so the policy is optimized based on the agent’s interaction with the environment. The mouse has to find the right balance between exploration and exploitation, i.e. between long and short term gain.
Reinforcement Learning is not only made for mice or mazes, but comes in very handy when we want the machine to learn to play the game of Go, or find the right advertisement to show to the user on YouTube to maximize overall gains. Simply put, whenever the machine is interacting with the environment and can improve its policy based on the outcomes.
This is just a brief (non-exhaustive) overview of some of the areas of machine learning and shows the rich tool set that is set at our disposal.
Thanks for reading! If you enjoyed this article, please leave a 👏
