支持向量机（SVM）中核函数的本质意义
本质上在做什么？
内积是距离度量，核函数相当于将低维空间的距离映射到高维空间的距离，并非对特征直接映射。
为什么要求核函数是对称且Gram矩阵是半正定？
核函数对应某一特征空间的内积，要求①核函数对称；②Gram矩阵半正定。
证明内积对应的Gram矩阵半正定：
α
T
K
α
=
[
α
1
,
α
2
,
⋯
,
α
n
]
[
κ
(
x
1
,
x
1
)
κ
(
x
1
,
x
2
)
⋯
κ
(
x
1
,
x
n
)
κ
(
x
2
,
x
1
)
κ
(
x
2
,
x
2
)
⋯
κ
(
x
1
,
x
n
)
⋮
⋮
⋱
⋮
κ
(
x
n
,
x
1
)
κ
(
x
n
,
x
2
)
⋯
κ
(
x
n
,
x
n
)
]
[
α
1
α
2
⋮
α
n
]
=
∑
i
=
1
n
∑
j
=
1
n
α
i
κ
(
x
i
,
x
j
)
α
j
=
∑
i
=
1
n
∑
j
=
1
n
α
i
α
j
⟨
ϕ
(
x
i
)
,
ϕ
(
x
j
)
⟩
=
⟨
∑
i
=
1
n
α
i
ϕ
(
x
i
)
,
∑
j
=
1
n
α
j
ϕ
(
x
j
)
⟩
=
∥
∑
i
=
1
n
α
i
ϕ
(
x
i
)
∥
2
2
⩾
0
\begin{aligned} {{ \bm \alpha}^{\rm T} {\bm K} { \bm \alpha}} &=\begin{bmatrix} {\alpha}_1, {\alpha}_2, \cdots, {\alpha}_n \end{bmatrix} \begin{bmatrix} \kappa \left( {\bm x}_1, {\bm x}_1 \right) &\kappa \left( {\bm x}_1, {\bm x}_2 \right) &\cdots &\kappa \left( {\bm x}_1, {\bm x}_n \right) \ \kappa \left( {\bm x}_2, {\bm x}_1 \right) &\kappa \left( {\bm x}_2, {\bm x}_2 \right) &\cdots &\kappa \left( {\bm x}_1, {\bm x}_n \right) \ \vdots &\vdots &\ddots &\vdots \ \kappa \left( {\bm x}_n, {\bm x}_1 \right) &\kappa \left( {\bm x}_n, {\bm x}_2 \right) &\cdots &\kappa \left( {\bm x}_n, {\bm x}_n \right) \ \end{bmatrix} \begin{bmatrix} {\alpha}_1 \ {\alpha}_2 \ \vdots \ {\alpha}_n \ \end{bmatrix} \ &= \sum\limits_{i=1}^{n} \sum\limits_{j=1}^{n} {\alpha}_i \kappa \left( {\bm x}_i, {\bm x}_j \right) {\alpha}_j \ &= \sum\limits_{i=1}^{n} \sum\limits_{j=1}^{n} {\alpha}_i {\alpha}_j \langle \phi \left( {\bm x}_i \right), \phi \left( {\bm x}_j \right) \rangle\ &= \langle \sum\limits_{i=1}^{n} {\alpha}_i \phi \left( {\bm x}_i \right), \sum\limits_{j=1}^{n} {\alpha}_j \phi \left( {\bm x}_j \right) \rangle \ &= \lVert \sum\limits_{i=1}^{n} {\alpha}_i \phi \left( {\bm x}_i \right) \rVert^2_2 \ &\geqslant 0 \end{aligned}
αTKα=[α1,α2,⋯,αn]
κ(x1,x1)κ(x2,x1)⋮κ(xn,x1)κ(x1,x2)κ(x2,x2)⋮κ(xn,x2)⋯⋯⋱⋯κ(x1,xn)κ(x1,xn)⋮κ(xn,xn)
α1α2⋮αn
=i=1∑nj=1∑nαiκ(xi,xj)αj=i=1∑nj=1∑nαiαj⟨ϕ(xi),ϕ(xj)⟩=⟨i=1∑nαiϕ(xi),j=1∑nαjϕ(xj)⟩=∥i=1∑nαiϕ(xi)∥22⩾0
