不确定成对约束的双对抗流形传播方法

成对约束传播(pairwise constraint propagation, PCP)通常研究的是在初始给定精确的成对约束基础上通过传播学习来增加成对约束的数量，从而给机器学习任务提供较多的监督信息。可是，在现实场景中，有时还有一些不精确的成对约束，因此，如何利用这些不精确的成对约束来提高成对约束传播学习的效果是一个有待解决的问题。针对这一问题，本文提出了一种不确定成对约束的传播方法。主要思想是用两个矩阵分别表示必须链接和不能链接的可能性，两种可能性之间形成对抗，同时两种成对约束之间也存在对抗关系，两类对抗相结合形成一种双对抗结构，作用于必须链接和不能链接的传播过程，使二者的对抗强度在竞争中趋于最小化。我们将该方法称为不确定成对约束传播(uncertain pairwise constraint propagation, UPCP)。在多个数据集上的实验结果表明，不确定成对约束的传播效果不超过但近似于理想化传播效果，在增强现实应用性的同时尽可能地保证了传播精度。

Doubly adversarial manifold propagation on uncertain pairwise constraints

Pairwise constraint propagation aims to increase the number of accurate pairwise constraints through propagation learning on the basis of the initial given exact pairwise constraints to provide additional supervision information for the machine learning task. However, inaccurate pairwise constraints occasionally exist in real scenes. Therefore, one problem to be solved is the use of these inaccurate pairwise constraints to enhance pairwise constraint propagation. A propagation method of uncertain pairwise constraints is proposed in this paper to solve this problem. The main idea is to use two matrices to represent the possibility of must-links and cannot-links. A confrontation between the two possibilities and a confrontation relationship between two kinds of pairwise constraints exist. The combination of two types of confrontation forms a double adversarial structure, which acts on the propagation process of must-links and cannot-links; therefore, the adversarial intensity between them is minimized in the competition. This method is called uncertain pairwise constraint propagation. Experimental results on multiple datasets show that the propagation effect of uncertain pairwise constraints does not exceed but is similar to the ideal propagation effect, which enhances the practical applicability and ensures the propagation accuracy completely.