Contrastive learning is a powerful self-supervised learning method, but we have a limited theoretical understanding of how it works and why it works. In this paper, we prove that contrastive learning with the standard InfoNCE loss is equivalent to spectral clustering on the similarity graph. Using...
2.1 Vertex Similarity 2.2 Contrastive Learning 2.3 Graph Pre-Training 3 GRAPH CONTRASTIVE CODING (GCC) 3.1 The GNN Pre-Training Problem 3.2 GCC Pre-Training 3.3 GCC Fine-Tuning 4 EXPERIMENTS 4.1 Pre-Training 4.2 Downstream Task Evaluation 4.2.1 Node Classification 4.2.2 Graph Classification 4.2.3 ...
-1) negatives = similarity_matrix[~masks].view(batch_size, -1)
To address the above issues, we propose CLAIS, a contrastive learning framework for graph-based vessel trajectory similarity computation. A combined parameterized trajectory augmentation scheme is proposed to generate similar trajectory sample pairs and a constructed spatial graph of the study re...
实际上这是一个contrastive learning框架哦,因为先天性支持图的关系,所以可以被用于self-supervised graph neural networks。但实际上用于Graph neural networks我们发现最有用的一个点是:看过我们ICLR2021 SSGC的朋友都记得SGC和SSGC这种方法虽然挺好用的,但是输入特征等于输出特征这点太坑爹了...来来来我们来看看如果...
patterns based on domain knowledge. the second line of research leverages the spectral graph theory to model structural similarity. in this work, we focus on structural similarity. unlike the above 2 genres, we adopt contrastive learning and graph neural networks to learn structural similarity from ...
本文的预训练任务:子图实例判别(subgraph instance discrimination)。对于每个顶点,将其 r-ego networks 作为采样子图实例,GCC 的目的是区分从特定顶点采样的子图和从其他顶点采样的子图。 2 Related work 接下来介绍几种顶点相似性: Neighborhood similarity
2) A consensus graph loss is proposed to approximate the consensus graph with all raw graphs so as to preserve the global spatial consistency of the learned graph; 3) Introduce a contrastive reconstruction loss to constrain the sample-level consistency, and to enhance the similarity between reconst...
2. 对比学习的基本概念 对比学习的核心思想是通过样本间的相似性与差异性来学习数据表示。传统的监督学习...
Due to the success observed in deep neural networks with contrastive learning, there has been a notable surge in research interest in graph contrastive learning, primarily attributed to its superior performance in graphs with limited labeled data. Within contrastive learning, the selection of a "view...