← AI Terminology
ML - Machine Learning
Machine learning is a subfield of AI in which systems learn from data — automatically improving their performance on a task through experience, without being explicitly programmed with rules for every case.
It encompasses supervised, unsupervised, and reinforcement learning paradigms.
It encompasses supervised, unsupervised, and reinforcement learning paradigms.
Why It Matters in AI
ML replaced hand-coded expert systems for tasks too complex to program explicitly: recognising faces, translating languages, detecting fraud, recommending content. The key insight — that statistical patterns in data can substitute for explicit rules — enabled AI to scale to the full complexity of the real world. Every modern AI product, from recommendation systems to LLMs, is built on ML foundations.
Key Points
| Aspect | Description |
|---|---|
| Unsupervised | Finds patterns without labels — clustering, dimensionality reduction, generative modelling |
| Deep learning | ML using multi-layer neural networks — the dominant paradigm since 2012 |
| Reinforcement | Agent learns via reward signals from interacting with an environment — games, robotics, RLHF |
| Generalisation | The core goal: learn patterns that transfer to unseen data, not just memorise training data |
| Traditional ML | Decision trees, SVMs, gradient boosting — still widely used for tabular data |
| Supervised learning | Learns from labelled examples (input, correct output) — classification, regression |
Simple Analogy
Teaching a child to recognise dogs: instead of defining "four legs, fur, tail…" (explicit programming), you show hundreds of photos labelled "dog" and "not dog." The child discovers the patterns themselves. Machine learning is this process applied to any data, at any scale, for any predictable task.
Common Usage Examples
sklearn.ensemble.RandomForestClassifier().fit(X_train, y_train)— traditional ML for tabular dataxgboost.XGBClassifier(n_estimators=300, max_depth=6)— gradient boosting, Kaggle standardtorch.nn.Linear(784, 10)— simplest neural network layer for MNIST digit classification- Recommendation: collaborative filtering learns user preferences from interaction history
- Fraud detection: ML classifies transactions as fraudulent based on historical labelled fraud cases
Summary
In short: Machine learning is the paradigm of teaching computers to learn from data rather than from explicit rules — the foundation of every modern AI application, from spam filters to large language models.