CS224W - Lecture 1.1 - Why Graphs

Last updated Aug 31, 2022

• On machine learning graphs
• Source course
• Status needs revision
• Type note

Metadata

• Source:: Stanford CS224W: Machine Learning with Graphs | 2021 | Lecture 1.1 - Why Graphs - YouTube
• Channel:: Stanford Online
• Presenter:: Jurij Leskovec
• Publish Date:: 2021-04-13
• Review Date::

1

 https://www.youtube.com/watch?v=JAB_plj2rbA

1

 00:32

• Graphs are a general language for describing and analyzing entities with relations/interactions.
• Many types of data can be represented as graphs:
  • Event Graphs
  • Computer Networks
  • Disease Pathways
  • Food Webs
  • Particle Networks
  • Underground Networks
  • Social Networks
  • Economic Networks
  • Communication Networks
  • Citation Networks
  • Internet
  • Network of Neurons
  • Knowledge Graphs
  • Regulatory Networks
  • Scene Graphs
  • Code Graphs
  • Molecules
  • 3D Shapes.

1

 03:10

• Networks or Natural Graphs
  • Underlying domains can naturally be represented as graphs.
  • Social Networks
  • Communication and transactions
  • Biomedicine
  • Brain Connections
• Graphs
  • Information/Knowledge
  • Software
  • Similarity networks
  • Relational structures
• Sometimes, the distinction between networks and graphs is blurred.

1

 05:26

• [?] How do we take advantage of relational structure for better prediction?
  • Complex domains have rich relational structure, which can be represented as a relational graph.
  • By explicitly modeling relationships, we achieve better performance.

1

 05:56

• Modern deep learning toolbox is designed for simple sequences and grids.
  • Such as images, audio, video, etc.

1

 06:35

• Graph/Networks are much harder to process because they are more complex.
  • They have arbitrary size and complex topological structure (i.e., no spatial locality like grids).
  • No fixed node ordering or reference point.
  • Often dynamic and have multimodal features.

1

 07:29

• Graphs are the new frontier of deep learning.

1

 08:23

• Representational learning allows us to automatically learn the features. We don’t have to perform manual feature engineering.
• To perform Representational Learning, we can map nodes to $d$-dimensional embeddings such that similar nodes in the network are embedded close together.

1

 10:04

• Traditional methods
• Methods for node embeddings
• Graph Neural Networks
• Knowledge graphs and reasoning
• Deep generative models for graphs
• Applications to Biomedicine, Science, Industry