无监督特征选择被认为是一个更难的问题,因为缺少引导相关信息搜索的类标记。
方差和可分离性
Variance score
暂且译为方差得分
可能是最简单的半监督特征评估
Laplacian Score
拉普拉斯得分
属于光谱特征选择家族
用于无监督特征选择
Lr=a˚i,j(fri−frj)2Sija˚i(fri−mr)2Dii L r = a ˚ i , j ( f r i − f r j ) 2 S i j a ˚ i ( f r i − m r ) 2 D i i
Cr=a˚(xi,xj)I^WML(fri−frj)2a˚(xi,xj)I^WCL(fri−firj)2 C r = a ˚ ( x i , x j ) I ^ W M L ( f r i − f r j ) 2 a ˚ ( x i , x j ) I ^ W C L ( f r i − f r j i ) 2
a˚i,j a ˚ i , j
argminv∈V arg min v ∈ V
I^ I ^
SPEC method
CSFS framework
Constraint Selection
aforementioned 前述的
nevertheless 虽然
depends merely on 仅仅依赖于
derogate 减损
coherent 相关
informativeness and coherence 信息量和连贯性
vicinity 附近
exert 施加
projected overlap 投影重叠
generalize 概括、归纳
joint 关联
dissimilarity 差异
Diagonal matrices 对角矩阵
algebraic steps 代数步骤
Cohd(W)=a˚mI^WML,cI^WCLd(overcm=0U`overmc=0)|WML||WCL| C o h d ( W ) = a ˚ m I ^ W M L , c I ^ W C L d ( o v e r c m = 0 U ` o v e r m c = 0 ) | W M L | | W C L |
a˚mI^WML,cI^WCLd(overcm=0U`overmc=0) a ˚ m I ^ W M L , c I ^ W C L d ( o v e r c m = 0 U ` o v e r m c = 0 )
|WML| | W M L | |WML||WCL| | W M L | | W C L |
SOM 自组织映射
Adaptive k-Neighborhood determination
低误差和更多保留局部的
Adaptive k-nearest-neighborhood 自适应k近邻用于聚类
Result
验证框架
Ascendant Hierarchical Clustering 上升的分层聚类
白血病、结肠癌数据集
和Laplacian, ReliefF,F2+r 4 and F3+r scores的比较
Results on Leukemia and Colon Cancer Data Sets
在白血病和结肠癌数据集的结果
Results on face images data sets
CONCLUSION
we proposed a framework for feature selection based on constraint selection for semi-supervised dimensionality reduction.
Future work may include the amelioration of the choice of labels set from which constraints are generated. Other perspectives may be the choice of clustering algorithm between the constraints.