How Machines Learn

Three Ways Machines Learn

7 min+60 XPStep 1/10

Three flavors of learning

All machine learning boils down to three setups — think of them as three different teachers:

Supervised — a teacher with an answer key. 'Here are 10,000 photos, each labeled cat or dog. Learn to tell them apart.'
Unsupervised — no answer key. 'Here's a pile of photos — find the patterns. Maybe some cluster together?'
Reinforcement — learn by trial and score. 'Try things; I'll give you points when it works.'

Same idea — learn from experience — three different kinds of feedback.

analogy

Supervised = flashcards with answers on the back. Unsupervised = sorting a junk drawer into groups that feel right. Reinforcement = learning to ride a bike — nobody tells you the physics; falling or not falling teaches you.