Skip to content
AI Atlas

Machine Learning

Letting programs infer rules from data instead of being hard-coded

OVERVIEW

All Concepts
6
Beginner
2
Intermediate
3
Expert
1

Machine learning compresses into a single sentence: given a set of samples, find a function that still works on new ones. Around that sentence grow the divide between supervised and unsupervised learning, the tug-of-war between overfitting and regularisation, and the methodology of measuring — honestly — whether it works at all.

Questions this domain answers

  • Q1

    What separates “memorising” from “learning”?

  • Q2

    Why does a model that aces training fail in production?

  • Q3

    What can unlabelled data still be used for?

Concepts in this domain

  1. 01Supervised LearningBeginnerPairs of questions and answers teach a model to answer on its own
  2. 02Unsupervised LearningBeginnerWith no answers given, structure must emerge from the data itself — and “good” has to be redefined
  3. 03Loss FunctionsIntermediateA loss function defines what you actually penalise — swap it and you swap your entire notion of right and wrong
  4. 04Overfitting & RegularizationIntermediateThe model memorises the training data word for word, then fails on anything new
  5. 05Model Evaluation & Cross-ValidationIntermediateAccuracy is the easiest metric to fool you — get evaluation wrong and everything else follows
  6. 06Bias–Variance TradeoffExpertEvery prediction error splits into three parts: the model too simple, the model too jumpy, and the world’s own randomness