Graph Convolutional Networks

1) Motivation, Objectives and Related Works:

Motivation:

• • The social network is best captured by a graph representation since pair-wise connection between two users do not form a grid. 

    1. Nodes of the graph represent users.

    2. Edges between two nodes represent connections between two nodes (users). 

    3. Each user has a three-dimensional feature matrix containing such as messages, images, and videos.

• • The connection between the structure and function of the brain to predict neural genetic diseases offers a motivational example to consider. 

    1. The brain is composed of several Region of Interest(s) (ROI). These ROIs is only locally connected to some surrounding regions of interests. 

    2. Adjacency matrix represents the degree of strengths between different regions of interest.

• • Quantum Chemistry also offers an interesting representation of graphical domain. 

    1. Each atom is represented by a node in graph and is connected to other atoms through bonds using edges. 

    2. Each of these molecules/atoms have different features, such as the associated charge, bond type etc.

Graph Definition and Characteristics:

• • Graph G is defined by:

    1. Vertices V

    2. Edges E

    3. Adjacency matrix A

  • Graph features:

    1. Node features: hi, hj (atom type)

    2. Edge features: eij (bond type)

    3. Graph features: g (molecule energy)

Objectives: 

2) Categories:

Paper List:

Reading List:

To-do List:

Unfilter:

3) Metrics:

References:

• https://atcold.github.io/pytorch-Deep-Learning/en/week13/13-1/

•