How I Understand the Different Types of Machine Learning as an ML Engineer
When I started learning Machine Learning, I came across many terms:
Supervised Learning.
Unsupervised Learning.
Semi-Supervised Learning.
Reinforcement Learning.
Initially, they felt like four definitions I simply had to memorize.
But as I continued learning Machine Learning, I realized something important:
Understanding how a model learns is more important than memorizing the names of algorithms.
So, I started looking at Machine Learning from a simpler perspective:
What information does the model have, and how does it learn from that information?
That question helped me understand the major types of Machine Learning much more clearly.
1. Supervised Learning — Learning With Answers
This was one of the first concepts I learned in Machine Learning.
The easiest way I understand supervised learning is:
The model learns from examples where the correct answer is already known.
For example, suppose I have data about students:
Hours Studied → Exam Score
2 → 45
4 → 60
6 → 72
8 → 85
Here, the model can learn the relationship between hours studied and exam score.
We give the model both:
Input → Output
The model learns the relationship and then uses it to make predictions on new data.
Classification
If the output is a category, we call it classification.
For example:
- Spam / Not Spam
- Fraud / Not Fraud
- Pass / Fail
- Cat / Dog
Some algorithms I have worked with or encountered include:
- Logistic Regression
- Decision Trees
- Random Forest
- KNN
- SVM
- Neural Networks
Regression
If the output is a numerical value, we call it regression.
For example:
- Predicting house prices
- Predicting sales
- Predicting temperature
- Predicting exam scores
Algorithms include:
- Linear Regression
- Ridge Regression
- Lasso Regression
- Decision Tree Regressor
- Random Forest Regressor
- Gradient Boosting
So my simple mental shortcut is:
Supervised Learning = Learning from data with known answers.
2. Unsupervised Learning — Finding Patterns
Then I came across Unsupervised Learning.
The biggest difference is simple:
There are no predefined answers.
Imagine giving someone information about thousands of customers without telling them which customers belong to which group.
Instead of providing labels such as premium customer, regular customer, or occasional customer, we ask the model to find patterns by itself.
This is where clustering becomes useful.
For example, an e-commerce company might discover groups of customers based on:
- Purchase frequency
- Spending amount
- Product preferences
- Shopping behavior
Algorithms such as:
- K-Means
- DBSCAN
- Hierarchical Clustering
can help identify these groups.
Another important concept here is dimensionality reduction.
When a dataset contains a large number of features, techniques such as PCA can help represent the data using fewer dimensions while retaining important information.
So I remember it as:
Unsupervised Learning = Finding patterns when the answers aren’t provided.
3. Semi-Supervised Learning — A Little Bit of Help
This one became easier for me when I thought about a real-world data problem.
Imagine having:
Get Gilnapradeep’s stories in your inbox
Join Medium for free to get updates from this writer.
100,000 images
but only:
5,000 images are labeled.
Manually labeling all 100,000 images could require a lot of time, effort, and domain expertise.
So what if we could use the small amount of labeled data together with the much larger amount of unlabeled data?
That’s the basic idea behind Semi-Supervised Learning.
It combines:
Labeled Data + Unlabeled Data
This approach can be useful in areas such as:
- Image classification
- Speech recognition
- Medical image analysis
- Document classification
The key idea I keep in mind is:
We don’t have enough labeled data, but we don’t want to waste the unlabeled data either.
4. Reinforcement Learning — Learning Through Feedback
Reinforcement Learning was a little different from the previous three concepts for me.
Instead of simply learning from a fixed dataset, the model learns through interaction.
The basic idea can be represented as:
Agent → Action → Environment → Reward → Learning
An agent interacts with an environment.
It takes an action.
The environment responds.
The agent receives a reward or penalty.
Over many interactions, the agent learns which actions can lead to better outcomes.
This idea is used in areas such as:
- Game-playing AI
- Robotics
- Autonomous systems
- Resource optimization
The simplest way I remember it is:
Reinforcement Learning = Learning through actions, feedback, and rewards.
The Mental Model I Use
After learning these concepts, I stopped trying to memorize long definitions.
Instead, I ask four questions.
Do I have labeled data?
Yes → Supervised Learning
Do I have unlabeled data and want to discover patterns?
Yes → Unsupervised Learning
Do I have a small amount of labeled data and a large amount of unlabeled data?
Yes → Semi-Supervised Learning
Does the system learn by interacting with an environment and receiving rewards?
Yes → Reinforcement Learning
This simple mental model makes the topic much easier for me to recall.
One Thing I’m Learning as an ML Engineer
One of the biggest lessons I’m taking from learning Machine Learning is that:
Choosing an algorithm should not be the first question. Understanding the problem and the data should be.
It’s easy to jump directly into questions like:
“Should I use Random Forest?”
“Should I use XGBoost?”
“Should I use an LSTM?”
But before choosing a model, I need to understand:
- What is the problem?
- What data do I have?
- Is the data labeled?
- What exactly am I trying to predict or discover?
- What kind of feedback is available?
- How will I evaluate the result?
Only after understanding these questions does choosing the appropriate learning approach start to make sense.
My Simple Summary
After working through these concepts, this is how I think about them:
Supervised Learning
→ I have the answers. Learn from them.
Unsupervised Learning
→ I don’t have the answers. Find the patterns.
Semi-Supervised Learning
→ I have some answers and lots of unanswered data.
Reinforcement Learning
→ Take actions, receive feedback, and improve.
Final Thought
Learning Machine Learning has taught me that technical concepts don’t always need complicated explanations.
Sometimes, finding the right mental model is enough to make a difficult concept simple.
For me, the four types of Machine Learning became much clearer once I stopped seeing them as definitions to memorize and started seeing them as different ways a machine can learn.
And I think that’s an important shift from learning Machine Learning to gradually thinking like an ML Engineer.
I’m still learning, experimenting, making mistakes, and connecting these concepts to real-world problems — but every time I understand one concept a little more deeply, the bigger picture of AI and Machine Learning becomes clearer.
What helped you understand the different types of Machine Learning when you were starting out?
#MachineLearning #AI #ArtificialIntelligence #MLEngineer #DataScience #Python #LearningInPublic #MachineLearningEngineer
That’s it for this part. More coming soon.
