Machine learning is part of AI. For modern AI, it is the engine, and the core idea is simple. Compared with old patterns of programming, humans gave strict rules for the systems to follow. In machine learning, the trick is that all the data, or most of the data, is given to the computer, and it is allowed to figure out the rules on its own.
An email app learning to spot spam from past examples, without anyone coding a rule for every spam message, is machine learning in action. It rests on three pillars.
Check Out: The AI/ML Programmes at Parul University.
First Pillar: Supervised Learning
AI has become an integral part of our lives. The details of it are necessary to understand. AI consists of Machine Learning (ML). Machine Learning has three pillars. One of them is supervised learning. In supervised learning, there is strict supervision. Imagine a strict teacher. And they are teaching a student, and data is fed to the computer. After being given the data, the computer or the system learns from it. The labelled data is used to anticipate new cases. Here we can observe that both the input and the right output is already with us or are known. Through this, the model learns the relationship and predicts new cases.
Bank loans: a bank feeds in thousands of past customer records showing who repaid (safe) and who defaulted (risky). Having learned from the correct answers, the model can predict whether a new applicant is a good risk.
Ticket pricing: past prices, with information on holidays, weekends and seasons, teach the model to price a ticket for tomorrow.
More examples: email spam detection, house-price prediction, student-result prediction, and disease diagnosis.
Pillar 2: Unsupervised Learning
As we move further, we get to know more about Machine Learning (ML). Unsupervised learning is the second pillar. Imagine a student who is learning, making notes, learning patterns, etc. on its own. No teacher is there to guide him/her. From this, we can infer that the system is given all the data, huge amounts of data. But there are no labels, correct or wrong answers, to decide which is which. The system has to find out on its own the hidden patterns. Here are some more examples:
Smart CCTV: a security system works to record the data. It watches thousands of hours of street videos. Here, no labels are given, whether it is human, animal, what is normal, etc. But over time, it reads and learns from the data, such as normal patterns of people, traffic, crossing roads, etc. If it finds any unusual thing, then AI flags it. Like someone crossing the red light or running suddenly.
Here are a few more examples to count on, like customer segmentation, market-basket analysis, grouping similar things, articles or images and fraud identification.
Pillar 3: Reinforcement Learning
The third pillar describes Reinforcement Learning. This is different from the previous ones. For example, you train a pet or learn how to play a video game. We observe here that a computer takes an action when it does well and gets rewarded for that. If the system fails to perform, a penalty is given too. This improves after millions of trials.
The self-driving car Tesla: the car is not hand-coded for every road situation, but it learns from the patterns and actions. Good actions and patterns get rewarded. When the car drives in the right lane, it gets rewarded, but drifting towards other cars or going off the road leads to a penalty. This is not achieved overnight; after many trials and errors and simulated actions, the car learns to drive.
Some more examples we can observe are robotics, warehouse automation, drone navigation, and game playing.
Also Read: How To Make Career in Machine Learning?
The Three Types at a Glance
In brief, we can say that:
Supervised learning is something where a teacher is there to guide the data; the data is already labelled to find out right or wrong. With unsupervised data, there are no teachers, but the system takes the unlabelled data and finds the answers through hidden patterns. The next one is reinforcement learning, where systems learn through rewards and penalties. To achieve the result, it has to go through a trial-and-error process.
Classification vs Regression
Within supervised learning, two problem types come up constantly. Classification predicts a category (spam or not spam, pass or fail, disease or no disease), the output is a label. Regression predicts a continuous number (house price, temperature, sales), the output is a value. Knowing which one your problem is decides which algorithm you reach for
How Do You Know a Model Is Good? Model Evaluation
Building a model is not enough; you must evaluate it before trusting it. Common metrics include accuracy, precision, recall, F1 score, and error rate, and the right one depends on the problem. A sharp example: for disease prediction, high recall (catching every real case) often matters more than raw accuracy, because missing a sick patient is far more costly than a false alarm. Choosing the right metric is a skill in itself, and students practise it hands-on, training and evaluating real models on the GPU-powered Lakshya 2047 labs.
FAQs
How to explain machine learning?
For a normal person to understand machine learning, it can be said that it is a part of AI. Data is given to a computer or system, and it learns from it in terms of how to operate it, rather than being specifically programmed. This is Machine Learning (ML). You give the system many examples, and it figures out the patterns itself, then uses them to make predictions, like learning to detect spam from past emails rather than being told every rule.
Are there three types of Machine Learning?
The three types are supervised learning (learning from labelled data with known answers), unsupervised learning (finding hidden patterns in unlabelled data), and reinforcement learning (learning through rewards and penalties by trial and error). Each suits different real-world problems.
What is the difference between classification and regression?
Both are supervised learning. Classification predicts a category or label (such as spam or not spam), so the output is categorical. Regression predicts a continuous numerical value (such as a house price), so the output is a number. The type of output decides which you use.
