Study Guide · Machine Learning · 5 min read

Machine Learning, Explained Simply

Normal programming is writing rules for a computer to follow. Machine learning flips it: show the computer thousands of examples, and let it work out the rules by itself.

Rules vs examples

Try writing rules to detect spam: 'block emails with FREE in caps' — spammers switch to FrEe. Every rule you write, attackers route around. The rulebook never catches up.

Instead, feed a learning algorithm ten thousand emails labeled spam or not-spam. It discovers the patterns itself — and when tactics change, you retrain on fresh examples rather than rewriting rules.

Learning, testing, generalizing

The crucial skill is not memorizing the examples but generalizing from them — performing well on emails it has never seen. That is why we hold back test data the model never trains on, as a final exam.

When a model aces training data but fails the exam, that is overfitting: memorization masquerading as learning. Nearly all of ML discipline exists to prevent this trap.

The three classic flavors

Supervised learning uses labeled examples (this is fraud, this is not). Unsupervised learning finds structure without labels, like grouping customers into segments. Reinforcement learning learns by trial, reward, and penalty.

Most enterprise value today still comes from supervised learning plus modern foundation models that learned language from vast unlabeled text.

Key Points

  • ML learns patterns from examples instead of hand-written rules.
  • Generalization — not memorization — is the whole point; held-out test data keeps models honest.
  • Overfitting is the field's central villain: great on training data, poor on reality.
  • Supervised, unsupervised, and reinforcement learning cover the main problem types.

All study guides for this term: Machine Learning, Explained Simply · How Machine Learning Works Under the Hood · Machine Learning in the Real World