当前位置：网站首页>Cs224w fall course - --- 1.1 why graphs?

Cs224w fall course - --- 1.1 why graphs?

2022-07-27 02:59:00 【chuanauc】

1. graph It's a kind of structed data

2. graph And picture （images） and written words Different ,graph There's no spatial locality

@ What is spatial locality ： In a string of text , We know the left and right of a character ; stay image in , We know that a grid is up, down, left and right in the space or other grids . however ,in network ,there is no reference point( Reference point ), namely , Undetermined network The node in graph Spatial location in .

No reference point also means in graph On , There is no fixed order to allow us to operate （ For example, text can be analyzed in chronological order from front to back text Node character of ）

3. In the past, supervised learning people spent a lot of time in finding the appropriate dimension of labeled data types , such as , Social networks should consider both the number of neighbors and the structure of neighbors, and so on , At that time feature engineering What we do is to analyze and judge the nodes , That attribute is important , Then, learn the... Of each node under this attribute dimension feature, That is, get the corresponding attribute embedding The embedded .

(embedding = representation = feature representation )

4. Course list ：

（1）"method to node embedding" This section will discuss Method for generating general node embedding （method to generate generic node embedding methods） , Such as deepwalk and Node2Vec

（2） We will study the expressive power of graph and why the above model of graph has this expressive power , What is the mechanism behind 【 After listening to the course, please come back and add the answer to this question by yourself 】

（3）how to scale models above up to large real graph

（4）Knowledge graphs and reasoning part ： Heterogeneous diagrams will be discussed 、 Knowledge map, etc

course arrangement ,10 Zhou 20 lesson ：