Machine learning has become a key tool for technological innovation and decision-making fueled by data. Personalized product recommendations in online stores, AI-assisted medical diagnosis systems, and predictive systems to forecast sales are all examples of machine learning’s presence in many aspects of our daily lives.

Notwithstanding, for those unfamiliar with the subject, the concepts and terms that constitute common machine learning jargon may sound complex. This article describes in a clear, simple, and precise manner the building blocks of machine learning and some of the most used algorithms to build systems that learn to make predictions or inference tasks from data.

What is Machine Learning?

Machine learning (ML) is a subarea of artificial intelligence (AI) that allows a computer system to learn from data to perform a specific task without explicitly being programmed with the instructions to do so. Instead, ML models are built during a process called training, in which they identify patterns and relationships in data to make predictions or inference-based decisions.

Areas of artificial intelligence and positioning of machine learning

To better understand this, ML is like getting skilled at cooking a recipe or a concrete type of cuisine. The more practice and experience, the better the skills to optimally adjust ingredients, cooking times, etc., yielding a more delicious result. Likewise, the more data a ML system is exposed to, the better it becomes at performing the task it is intended for. Data are the fuel with which ML systems acquire experience and become skilled.

The most common types of ML concepts, task types, and algorithms — procedural techniques followed by ML models to learn how to perform the task — are some of the notions demystified herein.

Five Fundamental Concepts

Supervised vs Unsupervised Learning: Most ML tasks can be broadly divided into supervised and unsupervised learning. In supervised learning, machines learn from data that already have “answers” or labels associated with data examples, e.g. categorized e-mails, and house listings with prices, and the goal is to be able to predict those labels for future unlabeled data examples. Meanwhile, in unsupervised learning data are not labeled, and the aim is to find patterns without knowledge of the “answers”.

Classification: A supervised learning task aimed at assigning categories to data, for instance classifying an email as spam or not, an animal species based on images, etc.

Regression: The numerical counterpart of classification. This supervised learning task consists in assigning or predicting continuous (numerical) values, given several data attributes. For instance, estimating the price of a house based on its properties is an example of a regression task.

Clustering: An unsupervised learning task whose goal is to group similar data into categories or clusters. A frequent use of clustering algorithms is the segmentation of customers with similar shopping habits.

Overfitting and Underfitting: These are frequent problems that ML might suffer, affecting their performance. Overfitting occurs when a model learns “too well” from the data used for training them. It can be deemed as memorizing the data, rather than acquiring the necessary knowledge to perform the task more generally, as a result of which the ML model performs poorly with new data not seen before. The opposite phenomenon is called underfitting, occurring when the model didn’t learn enough from the data — typically because the training data was insufficient — and doesn’t pick up on important aspects.

Four Fundamental Algorithms & Techniques

Linear Regression: One of the simplest ML algorithms, which tries to find an input-output linear relationship between data variables to predict a numerical output based on inputs. Predicting a house price based on its size, number of rooms, and age, is an example of problem solvable by linear regression.

Decision Trees: Used for both classification and regression tasks, these ML models look and operate like flowcharts, where nodes are questions leading towards an eventual decision at the bottom. They can be used for example to predict whether a customer will buy a product based on their age, behavior, and other features. Decision trees can be combined to solve more challenging predictive tasks where data are more complex; one of the most popular approaches to doing so is called random forests.

K-means: A well-known clustering algorithm to group data into categories based on similarity between examples.

Neural Networks: A more complex and sophisticated family of ML models for addressing more addressing tasks like object recognition in images. They work by mimicking the structure and information flow in animal brains, with layers of neurons that process and combine pieces of information to learn very complex patterns. Neural network models with a larger number of neuron layers are called deep neural networks. Deep learning is the subarea of ML focused on building these highly sophisticated models.

Curious to know what the ML models associated with these algorithms visually look like? The below image illustrates them:

Four types of machine learning algorithms and related models

Wrapping Up

Hopefully, if you are new to ML concepts and jargon, this article triggered your appetite for learning more about this exciting area of AI and advanced data analysis. Indeed, let me wrap up by outlining three real-world use cases that combine some of the ML tasks explained or apply more advanced ones:

• Personalized product and content recommendations in online platforms, based on the user needs and preferences
• Virtual assistants like Siri or Alexa, use ML together with NLP to understand human inquiries and reply adequately
• Facial recognition in smartphones, with underlying algorithms that learn your facial traits to unlock the device only when it’s you who appears in front of it