Two Central Questions
- Qa: Can we make sure Eout(g) is close enough to Ein(g) ?
(in: training data, out: testing data, g:hypothesis function) - Qb: Can we make Ein(g) small enough ?
Trade-off on M (Hypothesis Set)
- Small M
Qa: Yes, Qb: No (too few choices/ algorithms) - Large M
Qa: No, Qb: Yes - 找到合適的M,且了解為什麼無限的M是否可以學習 (就像前幾章PLA的問題),是接下來討討論的目標。
No comments:
Post a Comment