Understanding the basics of supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.
At the Canadian AI Advancement Institute (CAIAI), our mission is to help Canadians learn, adapt, and succeed with AI through accessible AI education and training. As AI systems become increasingly integrated into workplaces, educational institutions, and everyday life, it is essential for stakeholders to understand the basics of machine learning methods.
In this blog, we explore four foundational approaches to machine learning, including supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.
This blog provides an overview of four fundamental machine learning approaches that shape how modern AI systems learn and make decisions.
In summary, they can be described as follows:
Supervised learning is one of the most widely used and intuitive forms of machine learning. In this approach, models are trained using labelled datasets which contain collections of inputs paired with correct outputs. These labels serve as a “ground truth” that guides the learning process. As the model analyzes many examples, it begins to recognize patterns that enable it to predict on new, unseen data.
A common example is predictive text generation. Here, the input might be a sequence of words, and the output is the likely next word. By training on millions of labelled sequences, the model learns how language flows, enabling it to generate contextually appropriate suggestions. Another familiar application is email spam detection. In this case, messages categorized as “spam” or “not spam” form the labelled dataset. Over time, the model learns which patterns, such as phrasing, sender behaviour, or metadata, are more commonly associated with spam mail.
The strength of supervised learning lies in its precision. Because models receive explicit feedback during training, they continuously refine their outputs by minimizing error. However, this precision comes at a cost. Creating labelled datasets can be highly resource-intensive, often requiring expert human input.
In contrast to supervised learning, unsupervised learning does not rely on labelled data. Instead, models are tasked with exploring datasets independently, identifying hidden patterns, relationships, or structures within the data. Because there is no predefined output, these models uncover insights that might not be obvious to human analysts.
Common tasks undertaken by a model during unsupervised learning include:
Unsupervised learning also plays an important role in generative AI. When trained on large sets of unlabelled images, the AI model can learn the underlying distribution of shapes, textures, and features. They can then generate entirely new images that reflect these learned patterns.
Although this approach is powerful, evaluating its results can be difficult. Without labels, determining whether the model’s output is “correct” requires additional analysis or human interpretation. Nonetheless, unsupervised learning is indispensable for exploratory data analysis, anomaly detection, and building systems that adapt to new or evolving information.
The key difference between supervised and unsupervised learning lies in the presence or absence of labelled data. Supervised models learn from explicit examples, with correct answers provided at every step. In contrast, unsupervised models operate without labelled outputs, meaning they must infer structure on their own.
Their complementary strengths explain why organizations may employ both methods depending on their goals.
Semi-supervised learning combines the strengths of both supervised and unsupervised techniques. It is especially valuable in situations where labelled data is scarce or expensive, but unlabelled data is abundant. By using a small amount of labelled information to guide the model and incorporating larger unlabelled datasets to expand its learning, semi-supervised learning achieves strong performance at a lower cost.
Several strategies support this approach:
Semi-supervised learning is particularly effective in fields such as natural language processing, medical imaging, and fraud detection, where label acquisition is time-consuming or costly.
Reinforcement learning (RL) is fundamentally different from the other machine learning methods we have discussed so far. Rather than learning from labelled or unlabelled datasets, RL trains an agent through ongoing interaction with an environment. The agent takes actions, receives rewards or penalties, and adjusts its strategy to maximize cumulative reward.
This process is especially effective for creating task-specific agents. Classic examples include training AI to play chess or Go, where the agent refines its strategy through millions of simulated games. Over time, the agent learns which sequences of moves lead to successful outcomes.
Another example of the use of reinforcement learning here in Canada is RL Core, based in Edmonton, Alberta. The company emerged from the University of Alberta’s Reinforcement Learning and Artificial Intelligence Lab, widely recognized as home to many of the world’s top reinforcement learning experts.
RL Core specializes in applying reinforcement learning to manage industrial systems with adaptive, intelligent AI agents that continuously learn and optimize site performance in real time. Their software integrates with a site’s infrastructure, learns to optimize it, and continuously optimizes the site within the parameters it has been instructed to use.
So far, RL Core has been highly successful in managing water treatment, making a significant impact in Canada by optimizing the energy and chemicals required to treat our water.
For more information on this use case, visit rlcore.ai.
Reinforcement learning is also an increasingly important training method for consumer-facing generative AI tools (e.g., ChatGPT, Gemini, Claude, Perplexity), particularly via reinforcement learning from human feedback (RLHF). In this method, human evaluators score the model’s responses, helping shape its behaviour and alignment. This approach has been used to refine conversational AI systems, improving accuracy, safety, and usefulness.
As organizations continue to explore the benefits of AI, developing literacy in these machine learning approaches is an essential first step toward responsible and effective adoption.
For individuals and teams looking to deepen their knowledge, the Canadian AI Advancement Institute offers a range of opportunities, including AI education, an Introduction to Artificial Intelligence course, AI workshops, and AI training tailored to institutional, industry, and government needs.
Whether you are beginning your learning journey or preparing your organization for broader AI adoption, CAIAI offers a range of educational pathways to help you learn, adapt, and succeed with AI.
