机器学习基石上 (Machine Learning Foundations)—Mathematical Foundations
Hsuan-Tien Lin, 林轩田,副教授 (Associate Professor),资讯工程学系 (Computer Science and Information Engineering)
Training versus Testing
Recap and Preview
the Statistical Learning Flow
Two Central Questions
Tradeoff(权衡) on M (or |H|) M ( o r | H | )
M M can’t be
Effective Number of Lines
- 假设条件会有重叠的情况,除去 overlap,剩下多少种?
- must ≤2N ≤ 2 N
-
an=2+n(n−1)
a
n
=
2
+
n
(
n
−
1
)
- 化为平面上任意三点不共线的 n n 个点,用一条直线去划分,有多少种情况这个问题
- 直线把点分成两份,,分奇偶来证明
- 有递推式 an=an−1+(n−1) a n = a n − 1 + ( n − 1 ) ,设想新加的点和原来的一个点是一体的,有了 an−1 a n − 1 ,再用新点构造其他类别
- finite grouping of infinitely-many lines ∈H ∈ H ( H H 中的线的种类是有限的)
Effective Number of Hypotheses
dichotomies(mini-hypothesis) implemented by H H
dichotomy 数量的最大值 mH(N) m H ( N ) 来取代 M M <script type="math/tex" id="MathJax-Element-130">M</script>,growth function
Growth Function 不同类型
- Positive Rays
- Positive Interval
- Convex Sets
- Positive Rays
Break Point
polynomial(多项式的),exponential(指数的)
conjecture(猜想)