| layout | title | doodle | permalink |
|---|---|---|---|
page |
Graph Neural Networks |
/doodle.png |
/ |
Data science capstone domain of inquiry (DSC 180AB A13)
Developed by Gal Mishne and Aaron Fraenkel
- TOC {:toc}
This domain will investigate analysis of data lying on irregular structure, or graphs. This domain centers around understanding graph-based data and graphs as data. To approach this problem we will mainly be focusing on Graph Neural Networks.
Graphs and networks are playing an increasing role in modern machine learning and signal processing. We can view graph-based data as an augmented data structure, where in addition to having a feature representation for every datapoint (nodes in the graphs), we also have knowledge of the interactions and connectivity between points (graph edges). Graphs-based data is prevalent across domains such as biology (neuronal networks in the brain), computer science (social networks), electrical engineering (sensor networks), civil engineering (traffic networks) and many, many more.
Some questions we will be answering throughout this domain:
- How to model irregularly structured data?
- How to scale models and methods to large-scale graphs?
- How to benchmark machine learning approaches on graph-based data?
While a vast array of machine learning techniques rely on graphs and graph-based representations to analyze data (hierarchical clustering, spectral clustering, manifold learning, community detection, manifold/graph regularization), in this domain we will focus on geometric deep learning, or graph neural networks.
Participation in the weekly discussion section is mandatory. Each week, you are responsible for doing the reading/task assigned in the schedule. Come to section prepared to ask questions about and discuss the results of these tasks.
Each week, turn in answers to the weekly questions to Canvas. These questions are meant to focus your work for the week and help prepare you for discussion. If you have questions about your work, please ask them in section or office hours (I will rarely comment on your submission).
You are responsible for the entire weekly reading/task, even if portions are not covered in the weekly questions, as these are designed to help you in your replication. The weekly tasks are the building blocks for the project proposals/assignments due at the end of the quarter.
| Week | Topic |
|---|---|
| 1 | [Introduction]({{ "weeks/01-Introduction" |
| 2 | [Graph-based Data]({{ "/weeks/02-Data" |
| 3 | [Random walks on graphs]({{ "/weeks/03-Random-walks" |
| 4 | [Graph Embeddings]({{ "/weeks/04-Graph-Methods-II" |
| 5 | [Graph Convolutional Network]({{ "/weeks/05-GCN" |
| 6 | [Node Classification]({{ "/weeks/06-Node-Classification" |
| 7 | [Link Prediction]({{ "/weeks/07-Link-Prediction" |
| 8 | [Inductive Learning]({{ "/weeks/08-GraphSage" |
| 9 | [Recommender Systems]({{ "/weeks/09-recsys" |
| 10 | [Subgraph Learning]({{ "/weeks/10-subgraphs" |