W^=argminW∑i=1m||x(i)−∑j=1mwi,jx(j)||2
W^=argminW∑i=1m||x(i)−∑j=1mwi,jx(j)||2
wi,j=0 if i,j is not neighbor
wi,j=0 if i,j is not neighbor
∑j=1mwi,jx(j)=1
∑j=1mwi,jx(j)=1
第二步:将训练集降为d维(d < n),由于上一步已经求出局部特征矩阵W^W^,因此在d维空间也要尽可能符合这个矩阵W^W^,假设样本x(i)x(i)降维变为z(i)z(i),如下公式: