Pseudo Labeling is a technique used to improve the performance of
semi-supervised Graph Neural Networks (GNNs) by generating additional
pseudo-labels based on confident predictions. However, the quality of generated
pseudo-labels has long been a concern due to the sensitivity of the
classification objective to given labels. To avoid the untrustworthy
classification supervision indicating ``a node belongs to a specific class,''
we favor the fault-tolerant contrasting supervision demonstrating ``two nodes
do not belong to the same class.'' Thus, the problem of generating high-quality
pseudo-labels is then transformed into a relaxed version, i.e., finding
reliable contrasting pairs. To achieve this, we propose a general framework for
GNNs, termed Pseudo Contrastive Learning (PCL). It separates two nodes whose
positive and negative pseudo-labels target the same class. To incorporate
topological knowledge into learning, we devise a topologically weighted
contrastive loss that spends more effort separating negative pairs with smaller
topological distances. Additionally, to alleviate the heavy reliance on data
augmentation, we augment nodes only by applying dropout to the encoded
representations. Theoretically, we prove that PCL with the lightweight
augmentation works like a representation regularizer to effectively learn
separation between negative pairs. Experimentally, we employ PCL on various
models, which consistently outperform their counterparts using other popular
general techniques on five real-world graphs.

该论文提出了一种 Pseudo Contrastive Learning (PCL) 的通用框架，该框架通过生成可靠的对比对来解决生成高质量伪标签的问题，通过对负样本进行 dropout 来实现数据增强。 最终，实验证明该方法在五个真实世界的图上均优于其他普遍技术。

面向基于图的半监督学习的伪对比学习

Pseudo Contrastive Learning for Graph-based Semi-supervised Learning

For many biological image segmentation tasks, including topological
knowledge, such as the nesting of classes, can greatly improve results.
However, most `out-of-the-box' CNN models are still blind to such prior
information. In this paper, we propose a novel approach to encode this
information, through a multi-level activation layer and three compatible
losses. We benchmark all of them on nuclei segmentation in bright-field
microscopy cell images from the 2018 Data Science Bowl challenge, offering an
exemplary segmentation task with cells and nested subcellular structures. Our
scheme greatly speeds up learning, and outperforms standard multi-class
classification with soft-max activation and a previously proposed method
stemming from it, improving the Dice score significantly (p-values<0.007). Our
approach is conceptually simple, easy to implement and can be integrated in any
CNN architecture. It can be generalized to a higher number of classes, with or
without further relations of containment.

本研究提出一种新颖的方法，通过多层激活层和三种兼容的损失来编码基于拓扑学的先验信息，以解决生物图像分割任务中的问题，并在 2018 年的 Data Science Bowl 挑战中验证了该方法的效果，可以大幅提高分割的 Dice 分数。