「论文阅读」Reinforcement learning on graphs: A survey
论文名称：Reinforcement learning on graphs: A survey
作者：Nie Mingshuo, Chen Dongming, Wang Dongqi
时间：2022-4
出处：arxiv
代码：https://github.com/neunms/Reinforcement-Learning-on-Graph-A-Survey
摘要
图挖掘任务已经应用到许多领域中，比如社交网络，交通和电子商务等。这里有一些开创性的工作来，使用强化学习的技术来解决图数据挖掘任务。由于这些任务使用在不同领域中，这使得这些方法的比较存在一定的困难，在本文中，提供一个覆盖全局的视角来看RL和图挖掘方法，归纳这些图强化学习(GRL)为一个统一的公式。在本文中讨论了GRL方法的应用，简要描述，开源代码，基准数据集。并提出了在未来一些重要的方向和挑战。
介绍
RL的高速发展使得其被应用到多个跨学科领域中，近年来，学者们也开始考虑将RL域图挖掘进行结合，这激发了在使用有力的RL方法解决图挖掘任务中的决策问题上的兴趣。我们可以通过下图看到在2017~2022年发表的GRL文章数量： ...
「论文阅读」Graph Policy Network for Transferable Active Learning on Graphs
论文名称： Graph Policy Network for Transferable Active Learning on Graphs
作者：Shengding Hu, Zheng Xiong, Meng Qu, Xingdi Yuan, Marc-Alexandre Côté, Zhiyuan Liu, Jian Tang
期刊或会议：NeurIPS 2020
时间：2020-7
代码：ShengdingHu/GraphPolicyNetworkActiveLearning
原文摘要
Graph neural networks (GNNs) have been attracting increasing popularity due to their simplicity and effectiveness in a variety of fifields. However, a large number of labeled data is generally required to train these networks, which could be very e ...
「论文阅读(简)」Policy-GNN: Aggregation Optimization for Graph Neural Networks
论文名称：Policy-GNN: Aggregation Optimization for Graph Neural Networks
作者：Kwei-Herng Lai, Daochen Zha, Kaixiong Zhou, Xia Hu
时间：2020
期刊或会议：KDD '20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
代码：https://github.com/datamllab/Policy-GNN
原文摘要：
Graph data are pervasive in many real-world applications. Recently, increasing attention has been paid on graph neural networks (GNNs), which aim to model the local graph structures and capture the hierarchical patterns by aggregatin ...
「论文阅读(简)」Learning Robust Representations with Graph Denoising Policy Network
任务与存在问题
任务：图表征学习，目的是学习低维表示，以捕获在原始图上节点直接的依赖性。
存在问题：现有的表征学习方法是基于图神经网络，他们的变体依赖于邻居信息的聚合，这使得他们对图中的噪声非常敏感。
主要工作和创新点
提出了一个新颖的模型GDPNet，用于实现结合了强化学习的鲁棒性图表征学习。GDPNet由两部分组成，分别称为signal neighbor selection和representation learning。GDPNet有效地从噪声图数据中学习了节点的表征
将signal neighbor selection作为一个强化学习问题，使模型能够对待特定任务的奖励信号的弱监督下执行图去噪。
GDPNet能够生成不可见节点的表征 in an inductive fashion，其利用了图结构和节点特特征信息的联系。
证明了GDPNet在数学上等价于求解子模最大化问题，从而保证了模型可以得到有界的最优解。
提出的方法
Graph Denoising Policy Network(GDPNet)
根据signal neighbor选择环境的定义，本文引入了GDPNet模型，其 ...
图神经网络
视频学习笔记：https://www.bilibili.com/video/BV1334y1571g?spm_id_from=333.337.search-card.all.click
异构图
包含不同类型节点和连接
异构图的定义：节点类别的数量加上边的类别的数量大于2
元路径(Meta-path)：是连接两个对象的复合关系，是一种广泛使用的捕获语义的结构，比如，电影——演员——电影，这样的一个连接
基于元路径的邻居(Meta-path based Neighbors)：通过Meta-path找到的邻居信息
Heterogeneous Graph Attention Network（HAN）
这个算法主要包含了两个重要的计算方式，一个是节点级别的注意力(Node-level Attention)，另一个是语义级别的注意力(Semantic-level Attention)。
节点级别的注意力：对于原本给定的图结构，按照不同的元路径的定义，将其转化成另一种不同的图。举个例子，一张图的节点只有红和蓝两种颜色，定义一种元路径为红-蓝-红，然后根据这个元路径，利用基于元路径的邻居定义转化 ...
「论文阅读(简)」Vision GNN: An Image is Worth Graph of Nodes
期刊：Vision GNN: An Image is Worth Graph of Nodes
作者：Kai Han, Yunhe Wang, Jianyuan Guo, Yehui Tang, Enhua Wu
时间：2022. 7
期刊：arxiv
代码：https://github.com/huawei-noah/Efficient-AI-Backbones
原文摘要
Network architecture plays a key role in the deep learning-based computer vision system. The widely-used convolutional neural network and transformer treat the image as a grid or sequence structure, which is not flexible to capture irregular and complex objects. In this paper, we propose to represent the i ...
「论文阅读(简)」Graph Attention Network
论文名称：Graph Attention Networks
作者：Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, Yoshua Bengio
时间：2018
期刊或会议：ICLR
代码：https://github.com/Diego999/pyGAT
原文摘要
We present graph attention networks (GATs), novel neural network architectures that operate on graph-structured data, leveraging masked self-attentional layers to address the shortcomings of prior methods based on graph convolutions or their approximations. By stacking layers in which nodes are able to at ...
「论文阅读（简）」Semi-Supervised Classification with Graph Convolutional Network
论文名称：Semi-Supervised Classification with Graph Convolutional Network
作者：Kipf, Thomas N., Max Welling
时间：2017年
期刊或会议：ICLR
代码：https://github.com/dragen1860/GCN-PyTorch
原文摘要
We present a scalable approach for semi-supervised learning on graph-structured data that is based on an effificient variant of convolutional neural networks which operate directly on graphs. We motivate the choice of our convolutional architecture via a localized fifirst-order approximation of spectral graph convolutions. ...
「论文阅读(简)」Multi-VAE: Learning Disentangled View-common and View-peculiar Visual Representations for Multi-view Clustering
论文名称：Multi-VAE: Learning Disentangled View-common and View-peculiar Visual Representations for Multi-view Clustering
作者：Jie Xu; Yazhou Ren; Huayi Tang; Xiaorong Pu; Xiaofeng Zhu; Ming Zeng; Lifang He
年份：2021-10
期刊或会议：2021 IEEE/CVF International Conference on Computer Vision (ICCV)
代码：https://github.com/SubmissionsIn/Multi-VAE
原文摘要
Multi-view clustering, a long-standing and important research problem, focuses on mining complementary information from diverse views. However, existing works ofte ...
「论文阅读(简)」Reverse Graph Learning for Graph Neural Network
论文名称：Reverse Graph Learning for Graph Neural Network
作者：Liang Peng; Rongyao Hu; Fei Kong; Jiangzhang Gan; Yujie Mo; Xiaoshuang Shi; Xiaofeng Zhu
时间：2022-4
期刊：IEEE Transactions on Neural Networks and Learning Systems
原文摘要
Graph neural networks (GNNs) conduct feature learning by taking into account the local structure preservation of the data to produce discriminative features, but need to address the following issues, i.e., 1) the initial graph containing faulty and missing edges often affect ...
最新文章
