# Graph Neural Networks
###### tags: `機器學習`
因為網路上資源頗多,所以這邊就貼上看起來還不錯的學習資源當作參考。
reference: [YT 李弘毅 - GNN](https://www.youtube.com/watch?v=eybCCtNKwzA)
其他筆記:
* [番外.李宏毅学习笔记.12.GNN](https://blog.csdn.net/oldmao_2001/article/details/105457201)
* [【李宏毅2020 ML/DL】P18-19 Graph Neural Network](https://blog.csdn.net/weixin_42815609/article/details/107577956)
* [閱讀筆記 : A Comprehensive Survey on Graph Neural Networks ( Spatial-based ConvGNNs 篇)](https://z8663z.medium.com/%E9%96%B1%E8%AE%80%E7%AD%86%E8%A8%98-a-comprehensive-survey-on-graph-neural-networks-spatial-based-convgnns-%E7%AF%87-3aaeca24f371)
<br/>
## 術語補充:
1. **Aggregation:** 使用某 node neighbor feature update 下一層的hidden state。
2. **Readout:** 將所有 node 的 feature 集合成一個輸出。
( 因為一個 graph 通常就是要輸出一個結果嘛!例如: 一個分子是甚麼種類。)
<br/>
## 作法參考:
1. [NN4G ->Neural Network for Graphs: A Contextual Constructive Approach](https://ieeexplore.ieee.org/document/4773279)


2. DCNN(Dynamic Convolution Neural Network)
3. DGC (Dynamic Group Convolution)
4. MoNET (Mixture Model Network) - 事先定義好 weight

5. GAT (Graph Attension Network) - 自己去學鄰居的 weighted sum,對鄰居做 attension
6. GIN (Graph isomorphism Network)

影片: https://www.youtube.com/watch?v=M9ht8vsVEw8