概率保序语义哈希(POSH)算法:原理、扩展与实验评估
1. 优化算法概述
POSH的优化算法旨在获得最优离散哈希码 $B$。以下是算法的详细步骤:
Algorithm 4: Optimization Algorithm for POSH
Require: Data X ∈ℜd×n, the expected hash code length r; label matrix G ∈{0, 1}c×n;
maximum iterations Itern; parameters λ and η; iter = 0.
Ensure: Optimal discrete hashing codes B.
1: Initialize: Randomly initialize variables U, binary codes B and orthogonal matrix R.
2: while iter < Itern & not converge do
3:
U-Step: Update U by using Eq. (4.16) and update the auxiliary variable tii in Eq. (4.15);
4:
R-Step: Update R by using Eq. (4.21);
5:
B-Step: Update B by using Eq. (4.19).
6: end while
由于 $B$ 的离散约束,可推导出简化问题:
[
\max_{B} B^{\top} (R^{\top}G
超级会员免费看
订阅专栏 解锁全文

被折叠的 条评论
为什么被折叠?



