Skip to main content

Every Learner Is a Bet

This video presents the same text shown beside it, spoken and on screen. It adds nothing the text does not say.

State

With no assumptions, no learner beats any other averaged over all possible worlds — learning works only where a method's built-in bets fit the world it meets.

Show

Nearest neighbor bet on similarity; the tree on rectangles; the perceptron on one cut; convolution on nearby pixels; attention on computable relevance.

Watch for

The theorem stands behind every "what does it assume?" this course has asked.