Introduction to Graph Learning (Spring 2023)
Instructor: LI,Jia
TA: Gao, Ziqi
Time: We, Fr 4:30 PM - 5:50PM
Venue: E1-150
Introduction
Graph, as a very expressive model, has been widely used to model real-world entities and their relationships in application-specific networks. In this course, students will gain a thorough introduction to the basics of graph theories, as well as cutting-edge research in deep learning for graphs. The topics include graph embeddings, graph neural networks, graph clustering models, graph generative models, adversarial attacks on graphs, graph reasoning, etc.
Prerequisites
Linear Algebra (Strong)
Data Structure and Algorithms (Medium)
Probability theory (Medium) CS229
Announcements
April . 14 Next Wes (April. 19), there is no lecture due to business trip. To make up, we shall extend our lecture today (April. 14) and April. 21 to 4:30PM-6:30PM.
Mar . 24 Next Wes (Mar. 29), TA will give a tutorial about assignment and paper reading, the detailed requirement is here.
Mar . 17 we don't have lectures on Mar. 22. Instead, we invite Dr. Yu Rong to give a talk about geometric GNN at E1-101 1:30pm exactly that day.
Feb . 7 hi all. The first class starts at 4:30 pm, Feb. 8. See you.
Grading
Paper reading: 30% (must choose from the paper list, upload your paper reading video to Canvas. Due: Apr 22 Beijing Time)
Assignment: 20% (release: late Mar. Due: Apr 15)
Project: 50%
Reference and Handouts
Reference books and courses for extra reading:
[1] CS22W: Machine Learning with Graphs. Jure Leskovec.
[2] Computation & Machine Learning on Graphs. Pan Li.
[3] Graph Representation Learning. William l. Hamilton.
Handout 1: Introduction
Handout 2: Graph Embedding
Handout 3: Community Detection
Handout 4: Semi-Supervised Learning
Handout 5: Graph Neural Networks 1 [Tutorial]
Handout 6: Graph Neural Networks 2
Handout 7: GNN Training
Handout 8: Spatial-temporal GNN
Handout 9: Link Prediction
Handout 10: Graph-level Representation
Handout 11: Graph Self-supervised Learning
Handout 12: Knowledge Graph and ChatGPT
Paper Reading Schedule
The schedule is here: google doc. Book the topic (chosen from the following paper pool) as early as possible. Everyone's topic is unique. First come first serve!
Paper List
[1] DeepWalk. Brian Perozzi et al. KDD'14.
[2] node2vec. Aditya Grover et al. KDD'16.
[3] GCN. Thomas N. Kipf et al. ICLR'17.
[4] GAT. Petar Veličković et al. ICLR'18.
[5] GIN. Keyulu Xu et al. ICLR'19.
[6] SGC. Felix Wu et al. ICML'19.
[7] DGI. Petar Veličković et al. ICLR'19.
[8] GMI. Zhen Peng et al. WWW'20.
[9] DCRNN. Yaguang Li et al. ICLR'18.
[10] STGCN. Bing Yu et al. IJCAI'18.
[11] Nettack. Daniel Zügner et al. KDD'18.
[12] CD-ATTACK. Jia Li et al. WWW'20.
[13] DiffPool. Zhitao Ying et al. NeurIPS'18.
[14] MinCutPool. Filippo Maria Bianchi et al. ICML'20.
[15] GraphRNN. Jiaxuan You et al. ICML'18.
[16] JT-VAE. Wengong Jin et al. ICML'18.
[17] GCMC. Rianne van den Berg et al. KDD'18.
[18] PinSAGE. Zhitao Ying et al. KDD'18.
[19] TransE. Antoine Bordes et al. NeurIPS'13.
[20] GMN. Yujia Li et al. ICML'19.
[21] SEAL. Muhan Zhang et al. NeurIPS'18.
[22] LDS. Luca Franceschi et al. ICML'19.
Project
Each one chooses a research topic related to course material, e.g., graph learning. The report should follow ACM format with strict 6 pages limitation, including reference and appendix, see the following for reference https://kdd.org/kdd2021/calls/view/call-for-research-track-papers. Here are some tips:
The report should at least consist of introduction, related work, methodology and experiment. Theoretical deviation is not a necessity but encouraged.
Use concise and clear language.
Clearly declare your difference with previous works.
If there is any theoretical deviation, check your assumption and make sure it is non-fragile.
The ddl for the final report is May 25th Beijing time, to the email of TA ( zgaoat@connect.ust.hk ).
Resources
Understanding Matrix
Machine Learning
A few useful things to know about machine learning
Software
DGL (an open-source library for deep learning on graphs)
Graph learning tutorial
Advanced deep graph learning: deeper, faster, robuster and unsupervised (WWW'21 tutorial)
Graph learning publications