Graph Transfer Learning

Graph embeddings have been tremendously successful at producing node representations that are discriminative for downstream tasks. In this paper, we study the problem of graph transfer learning: given two graphs and labels in the nodes of the first graph, we wish to predict the labels on the second...

Full description

Saved in:
Bibliographic Details
Main Authors: Gritsenko, Andrey, Guo, Yuan, Shayestehfard, Kimia, Moharrer, Armin, Dy, Jennifer, Ioannidis, Stratis
Format: Conference Proceeding
Language:English
Subjects:
Conferences
Data mining
Epidemics
graph embedding
Task analysis
Training
Transfer learning
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Graph embeddings have been tremendously successful at producing node representations that are discriminative for downstream tasks. In this paper, we study the problem of graph transfer learning: given two graphs and labels in the nodes of the first graph, we wish to predict the labels on the second graph. We propose a tractable, non-combinatorial method for solving the graph transfer learning problem by combining classification and embedding losses with a continuous, convex penalty motivated by tractable graph distances. We demonstrate that our method successfully predicts labels across graphs with almost perfect accuracy; in the same scenarios, training embeddings through standard methods leads to predictions that are no better than random.
ISSN:2374-8486
DOI:10.1109/ICDM51629.2021.00024