Types of machine learning
Broadly speaking, there are three paradigms of machine learning: supervised learning, unsupervised learning, and reinforcement learning.2
Supervised learning In supervised learning, the algorithm uses existing information or data points that come with clear labels or categories.2 For example, the algorithm is trained on a number of cat and dog pictures (of various types) to differentiate between the two animals based on features like color, size, ears, nose, etc. Once it is trained, it can then label any new images given to it as “cat” or “dog”.
This enables the machine to identify dogs and cats of various colors, sizes, and features.
The objective of supervised learning is for the machine to learn from many labeled examples.
Unsupervised learning In unsupervised learning, the data points exist without any attached labels or categories. Unlike supervised learning, which deals with data that is already labeled, unsupervised learning focuses on organizing data and uncovering hidden patterns.2
The unsupervised learning method involves organizing data into distinct clusters or exploring different viewpoints to simplify complex data and enhance data analysis.3 The goal is to discover meaningful connections within the data rather than relying on predefined labels.
While supervised learning assigns labels like “cat” and “dog” to images, unsupervised learning involves finding inherent patterns without the aid of explicit labels. Imagine a machine learning system is tasked with sorting through a collection of wildlife photographs that don’t include information on animal species. In this example, unsupervised learning algorithms can identify similar features in images and group them without being explicitly instructed about the specific animal.
Reinforcement learning In reinforcement learning, the algorithm determines the best course of action based on data and feedback.2 For example, a robot uses its sensors to process information at certain time intervals. Once the algorithm categorizes and labels the collected information, an immediate feedback signal evaluates the algorithm’s effectiveness.
This iterative process enables the algorithm to refine the labeling process over time. The algorithm processes input data to create outputs, followed by an immediate feedback signal that gauges the effectiveness of the output. Based on this signal, the algorithm adjusts itself in order to achieve the highest reward.
2, 3: See the Bibliography in this module's Summary section.