图机器学习基础知识——CS224W（01-Intro）

CS224W: Machine Learning with Graphs

Stanford / Winter 2021

01-Intro

• Why Graph Machine Learning So Hard ?

  • Arbitrary size and complex topological structure (No spatial locality like grids)

  • No fixed node ordering or reference point

  • Often dynamic and have multimodal features

  

• CS224W & Representation Learning

  • (Supervised) Machine Learning Lifecycle: This feature, that feature. Every single time!

  • Map nodes to d-dimensional embeddings such that similar nodes in the network are embedded close together

  

• Different Types of Tasks

  • Node level

  • Edge-level

  • Community (subgraph) level

  • Graph-level prediction, Graph generation

  

• Classic Graph ML Tasks

  • Node classification: Predict a property of a node

    • Categorize online users/items

  • Link prediction: Predict whether there are missing links between two nodes

    • Knowledge graph completion

  • Graph classification: Categorize different graphs

  • Clustering: Detect if nodes from a community

    • Social circle detection

  • Other tasks

    • Graph generation: Drug discovery

    • Graph evolution: Physical simulation

• Compoments of a Network

  • Objects: nodes, vertices $N$

  • Interations: links, edges $E$

  • System: network, graph $G(N,E)$

• How do you define a graph ?

  • The way you assign links will determine the nature of the question you can study

• Directed vs. Undirected Graphs

• Node Degrees

• Bipartite Graph

  二部图

  • A graph whose nodes can be divided into two disjoint sets $U$ and $V$ such that every link connects a node in $U$ to one in $V$; that is, $U$ and $V$ are independent sets

  • Authors-to-Papers (they authored)

  • Actors-to-Movies (they appeared in)

  

• Folded/Projected Bipartite Graphs

  • Create a connection between a pair of nodes if they have at least one neighbor in common

  

• Representing Graphs: Adjacency Matrix

  • Adjacency Matrices are Sparse

  • Networks are Sparse Graphs: Most real-world networks are sparse ($E<<E_{max}$ or $k<<N-1$)

• More Types of Graphs

  • Unweighted and Weighted

    

  • Self-edges (self-loops) and Multigraph

    

• Connectivity of Undirected Graphs

  无向图的连通性

  • Connected (undirected) graph: Any two vertices can be joined by a path

  • Disconnected graph: Made up by two or more connected components
    

  • 无向非连通图的邻接矩阵可以被写成分块对角矩阵的形式（block-diagonal form），因为零元素可以被限制在次对角的两个分块中

    

• Connectivity of Directed Graphs

  有向图连通性

  • Strongly connected directed graph: has a path from each node to every other node and vice versa (A-B and B-A)

  • Weakly connected directed graph: is (weakly) connected if we disregard the edge directions (e.g. its corresponding undirected graph is connected)

    

  • Strongly connected components (SCCs) can be identified, but not every node is part of a nontrivial strongly connected component

    

Supplement From Graph Representation Learning

• Homogeneous and Heterogeneous Graph

  同质图和异质图

  • Homogeneous Graph: 节点和边的类型（关系类型）只有一种

  • Heterogeneous Graph：节点和边的类型（关系类型）不止一种

• Graph or Network ?

  • Graph: 抽象数据结构，更关注数学层面、抽象层面的理论性质

  • Network: 抽象数据结构的实例化，更关注真实数据本身的特征与性质

• 节点分类中，训练数据（图中的节点）打破了i.i.d.的假设，对一组相互连接的节点进行建模，而不是对一组符合i.i.d.的节点进行建模（但在Graph-level的任务中，训练数据是整张图，符合i.i.d.）

• Link Prediction = Relational Inference