Cousera机器学习基石第六周笔记 Machine Learning Foundation Week 6 Note in Cousera
Theory of Generalization
test error can approximate training error if there is enough data and growth function does not grow too fast
Restriction of Breaking Point
The Four Breaking Points
growth function \(m_\H(N)\):max number of dichotomies
- positive rays: \(m_\H(N)=N+1\)
- positive intervals: \(m_\H(N)=\frac{1}{2}N^2+\frac{1}{2}N+1\)
- convex sets: \(m_\H(N)=2^N\)
- 2D perceptrons :\(m_\H(N)<2^N\) in some cases
Restriction of Breaking Point
