Generalizing graph neural networks across graphs, time, and tasks

Graph-structured data are ubiquitous across numerous real-world contexts, encompassing social networks, commercial graphs, bibliographic networks, and biological systems. Delving into the analysis of these graphs can yield significant understanding pertaining to their corresponding application field...

Full description

Saved in:
Bibliographic Details
Main Author: WEN, Zhihao
Format: text
Language:English
Published: Institutional Knowledge at Singapore Management University 2023
Subjects:
GNN
Online Access:https://ink.library.smu.edu.sg/etd_coll/507
https://ink.library.smu.edu.sg/context/etd_coll/article/1505/viewcontent/GPIS_AY2019_PhD_WEN_Zhihao.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Singapore Management University
Language: English
Description
Summary:Graph-structured data are ubiquitous across numerous real-world contexts, encompassing social networks, commercial graphs, bibliographic networks, and biological systems. Delving into the analysis of these graphs can yield significant understanding pertaining to their corresponding application fields.Graph representation learning offers a potent solution to graph analytics challenges by transforming a graph into a low-dimensional space while preserving its information to the greatest extent possible. This conversion into low-dimensional vectors enables the efficient computation of subsequent graph algorithms. The majority of prior research has concentrated on deriving node representations from a single, static graph. However, numerous real-world situations demand rapid generation of representations for previously unencountered nodes, novel edges, or entirely new graphs. This inductive capability is vital for highperformance machine learning systems that operate on ever-changing graphs and consistently encounter unfamiliar nodes. The inductive graph representation presents considerable difficulty when compared to the transductive setting, as it necessitates the alignment of new subgraphs containing previously unseen nodes with an already trained neural network. We further investigate inductive graph representation learning through three distinct angles: (1) Generalizing Graph Neural Networks (GNNs) across graphs, addressing semi-supervised node classification across multiple graphs; (2) Generalizing GNNs across time, focusing on temporal link prediction; and (3) Generalizing GNNs across tasks, tackling various low-resource text classification tasks.