问题
如果有2个矩阵 Y ∈ R n × q Y \in R^{n \times q} Y∈Rn×q 和 Z ∈ R n × c Z \in R^{n \times c} Z∈Rn×c ,如何度量两个矩阵的相似性?
回答
采用The Hilbert-Schmidt Independence Criterion (HSIC)进行度量。
H
S
I
C
(
Z
,
Y
)
=
(
n
−
1
)
−
2
T
r
(
H
K
H
Q
)
HSIC(Z, Y) = (n − 1)^{-2}Tr(HKHQ)
HSIC(Z,Y)=(n−1)−2Tr(HKHQ)
其中,
K
=
Z
Z
T
K = ZZ^T
K=ZZT and
Q
=
Y
Y
T
Q = YY^T
Q=YYT are the inner
product of instances in Z and Y respectively and
H
=
I
−
1
n
e
e
T
H = I − \frac{1}{n} ee^T
H=I−n1eeT and
I
I
I denote the identity matrix(单位矩阵),and
e
e
e be a vector of all ones.
理解
- 计算结果值越大,表示越接近。
参考:
[1] Huang R, Wu Z. Multi-label feature selection via manifold regularization and dependence maximization[J]. Pattern Recognition, 2021, 120: 108149.
[2] Gretton A , Bousquet O , Smola A , et al. Measuring Statistical Dependence with Hilbert-Schmidt Norms[J]. Springer, Berlin, Heidelberg, 2005.