ManySpace
Türkçe Sign in Sign up

AI & ML

Learning Theory

Part of Foundations, level 1. About 5 hours of study across the topics below.

Foundations

  1. Inductive Bias and the No-Free-Lunch PrincipleEvery learner prefers some hypotheses; without assumptions no learner beats another across all problems.1–2 h

Pac

  1. PAC LearningProbably approximately correct: how many samples guarantee small error with high confidence for a finite hypothesis class.1–2 h
  2. VC DimensionMeasuring capacity of infinite hypothesis classes by the largest point set they can shatter.1–2 h

Online

  1. Online Learning and RegretLearning from a stream against an adversary: regret as the performance measure, multiplicative weights as the algorithm.1–2 h

Modern Generalization

  1. Double Descent and Implicit RegularizationWhy heavily overparameterized networks can generalize: test error that falls again past the interpolation threshold.1–2 h