Graph Deep Learning Based on Graph Convolutional Networks

Summary: The leader in processing graph-structured data!

For more in-depth articles, please pay attention to the cloud computing channel:

https://yq.aliyun.com/cloud
        Graph deep learning based on graph convolutional network
screenshot
Let’s briefly review what deep learning has done successfully!
screenshotDeep
learning has made great achievements in speech recognition, image recognition, natural speech processing and other fields in recent years. ImageNet: It is a computer vision system recognition project, which is currently the largest database of image recognition in the world, and is well known in the industry.
Let’s first review the European-style deep learning technology without big data support. For the identification of a letter "Z", we usually build a 2D grid (dot matrix), if the points in it are connected, the definition of such a connection is formed by "Z". Then it is tested with other letters, the correctness of this model.

screenshot
screenshot
The traditional deep learning method is actually a process of manually designing features. Also, there is no guarantee of accuracy. In real deep learning, end-to-end learning, the designer does not know what happened in the process, and naturally he will not interfere artificially.
screenshot
screenshot
If the data cannot be gridded, then CNNs are useless. Therefore, CNNs still have many defects to a certain extent. For example, graph structure data, how to deal with it? There are many examples of this in the real world: social networks (the famous Six Degrees Theory), World Wide Web, knowledge graphs, etc. These are graph structures, not grid structures, how do we solve these.
screenshot
screenshot
Here's a simple solution to graph-structured data.
What kind of problems will there be in the screenshot

method? What do we need to solve the problem?
screenshot
Let’s briefly introduce the GCNs of first-order message passing. This theory has been proposed in 2009.
screenshotNext
, let's take a look at the GCN model architecture!
screenshot
What can the GCN model architecture do? Take a small chestnut first.
screenshot
What is the relationship between the GCN model and the famous Weylerhmann algorithm? Semi-supervised classification of screenshots
is also a good approach. screenshot Semi-supervised classification embedding method - two-step pipeline, this method also has some problems, but I think it can be solved. The screenshot is a small chestnut. The video link is a small example about semi-supervised learning. Interested friends can take a look. Screenshot video: http://tkipf.github.io/graph-convolutional-networks/ In addition, there is a classification of citation networks, which can also be implemented by this method. screenshotThe experimental results of the 2-layerGCN model below screenshot














The screenshot
also includes some recent examples of this method applied to other programs.
screenshot
Use this method to link predictions about graph auto-encoders.
Below is the introduction screenshot of auto-encoders
Autoencoders
Further reading
Blog post Graph Convolutional Networks:
http://tkipf.github.io/graph-convolutional-networks
Code on Github:
http://github.com/tkipf/gcn
Kipf & Welling, Semi-Supervised Classification with Graph Convolutional Networks, ICLR 2017:
https://arxiv.org/abs/1609.02907
Kipf & Welling, Variational Graph Auto-Encoders, NIPS BDL Workshop, 2016: https://arxiv.org/abs /1611.07308
screenshot
Author: Thomas
Kipf

The original title of the article is "Deep Learning on Graphs with Graph Convolutional Networks", author: Thomas Kipf, translator: Yuan Hu, reviewer: I am the brother of the theme song, the attachment is the pdf of the original text.
The article is a simplified translation, for more details, please check the original text