Semi-Supervised Classification with Graph Convolutional Networks
By Thomas Kipf, Max Welling. International Conference on Learning Representations, 2016
36,100 citations recorded by the source corpus.
We present a scalable approach for semi-supervised learning on graph-structured data that is based on an efficient variant of convolutional neural networks which operate directly on graphs. We motivate the choice of our convolutional architecture via a localized first-order approximation of spectral graph convolutions. Our model scales linearly in the number of graph edges and learns hidden layer representations that encode both local graph structure and features of nodes. In a number of experiments on citation networks and on a knowledge graph dataset we demonstrate that our approach outperforms related methods by a significant margin.
Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Graph Learning
View Paper: https://arxiv.org/abs/1609.02907
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohit Bansal
- Yoshua Bengio
- Jianfeng Gao
- Percy Liang
- Quoc V. Le
- Sergey Levine
- Ruslan Salakhutdinov
- Dacheng Tao
In Representation & Self-Supervised Learning
- Yoshua Bengio
- Trevor Darrell
- Dacheng Tao
- William L. Hamilton
- Han-Jia Ye
- Fei Sha
- Jure Leskovec
- Andrea Vedaldi
In Graph Learning
- William L. Hamilton
- Jure Leskovec
- Stephan Günnemann
- Matthias Fey
- Michael M. Bronstein
- Rex Ying
- Jian Tang
- Aleksandar Bojchevski
Citations in the corpus
- Graph Attention Networks (2017, methodology): “Transductive learning For transductive learning tasks, we compare against the same strong base-lines and state-of-the-art approaches as specified in Kipf & Welling (2017).”
- Inductive Representation Learning on Large Graphs (2017, positive): “Since, the “convolutional” variant of GraphSAGE is an extended, inductive version of Kipf et al’s semi-supervised GCN [17], we term this variant GraphSAGE-GCN.”
- Unsupervised Data Augmentation for Consistency Training (2019, methodology): “On this direction, semisupervised learning [4] is one of the most promising methods and recent works can be grouped into three categories: (1) graph-based label propagation via graph convolution [31] and graph embeddings [62], (2) modeling prediction target as latent variables [30], and (3) consiste”
- Reasoning Over Semantic-Level Graph for Fact Checking (2019, methodology): “To capture the multi-hop relational information, we employ multi-layer graph convolutional network (GCNs) Graph-based Reasoning 12 claim … [SEP] … sentence1 … sentence2 … (Kipf and Welling 2016) to update the node representation.”
- Object-Centric Learning with Slot Attention (2020, neutral): “Graph neural networks [54–57] and in particular the self-attention mechanism of the Transformer model [20] are frequently used to transform sets of elements with constant cardinality (i.”
- Self-Training With Noisy Student Improves ImageNet Classification (2019, methodology): “Finally, frameworks in semi-supervised learning also include graph-based methods [84, 73, 77, 33], methods that make use of latent variables as target variables [32, 42, 78] and methods based on low-density separation [21, 58, 15], which might provide complementary benefits to our method.”
- A simple neural network module for relational reasoning (2017, background): “There is recent work applying neural networks to graphs, which are a natural structure for formalising relations [12, 19, 33, 37, 26, 2].”
- Geometric Deep Learning on Graphs and Manifolds Using Mixture Model CNNs (2016, methodology): “Kipf and Welling [21] further simplified this approach using simple filters operating on 1-hop neighborhoods of the graph.”
- Stochastic Training of Graph Convolutional Networks (2017, methodology): “This is significant because most GCNs only have L = 2 layers (Kipf & Welling, 2017; Hamilton et al., 2017a).”
- Variational Graph Auto-Encoders (2016, methodology): “We demonstrate this model using a graph convolutional network (GCN) [4] encoder and a simple inner product decoder.”
- Neural Logic Machines (2019, background): “The learning system should deal with high-order relational data and quantifiers, which goes beyond the scope of typical graph-structured neural networks (Kipf & Welling, 2017).”
- Structured Convolutional Kernel Networks for Airline Crew Scheduling (2021, methodology): “Finally, note that in recent papers, convolutional graph neural networks (ConvGNNs) (Kipf & Welling, 2016) are used either (1) to warm-start a solver (trained under the imitation learning framework) (Owerko et al., 2020), (2) to solve the optimization problem end-to-end (Khalil et al., 2017), or…”
- Knowledge Graph Convolutional Networks for Recommender Systems (2019, background): “[10] propose a convolutional architecture via a localized first-order approximation of spectral graph convolutions.”
- Neural Message Passing for Quantum Chemistry (2017, methodology): “The Kipf & Welling (2016) model results in a message function Mt(htv, h t w) = cvwh t w where cvw = (deg(v)deg(w))−1/2Avw.”
- Strategies for Pre-training Graph Neural Networks (2019, methodology): “In the first category are methods that use local random walk-based objectives (Grover & Leskovec, 2016; Perozzi et al., 2014; Tang et al., 2015) and methods that reconstruct a graph’s adjacency matrix, e.g., by predicting edge existence (Hamilton et al., 2017a; Kipf & Welling, 2016).”
- E(n) Equivariant Graph Neural Networks (2021, methodology): “Similarly to the work of (Kipf & Welling, 2016b) further extended by section 3.3 in (Liu et al., 2019), our graph auto-encoder z = q ( G ) embeds a graph G into a set of latent nodes z = { z 1 , . . . z M } ∈ R M × n , where M is the number of nodes and n the embedding size per node.”
- Directional Message Passing for Molecular Graphs (2020, background): “General GNNs have been largely inspired by their application to molecular graphs and have started to achieve breakthrough performance in various tasks at around the same time the molecular variants did (Kipf & Welling, 2017; Gasteiger et al., 2019; Zambaldi et al., 2019).”