Types of Machine Learning
Not all learning is the same. Just as humans learn from teachers, exploration, and trial-and-error, ML has three distinct paradigms — each suited to different problems.
Supervised Learning
The most widely used paradigm. You provide labeled training data — input-output pairs — and the model learns to predict outputs for new inputs.
Classification vs Regression
Supervised learning splits into two tasks depending on the type of output.
Supervised Toolbox
Each algorithm has its sweet spot — from simple to powerful.
- Linear Regression — simple, interpretable baselines
- Decision Trees — handle non-linear data naturally
- SVMs — optimal decision boundaries
- Neural Networks — learn complex mappings at scale
Unsupervised Learning
No labels, no teacher. The algorithm discovers hidden structure in data entirely on its own.
K-Means Clustering
Group similar data points by minimizing the distance to cluster centers.
Reduction & Anomalies
Compress data and catch outliers — two more unsupervised superpowers.
- PCA — find directions of max variance
- Autoencoders — neural compression
- Anomaly Detection — flag fraud, defects, intrusions
Reinforcement Learning
An agent interacts with an environment, takes actions, and learns a policy that maximizes cumulative reward.
The Return
The agent maximizes the discounted cumulative reward over time.
Beyond the Three
Modern AI blurs the boundaries with hybrid learning strategies.
Choosing the Right Type
Match your data and problem to the right paradigm.
- Have labels? → Supervised learning
- Find structure? → Unsupervised learning
- Sequential decisions? → Reinforcement learning
- Few labels? → Semi/self-supervised
What you learned
Three paradigms — supervised, unsupervised, and reinforcement — plus hybrid approaches. Choosing the right one depends on your data and your problem.