Abstract

The ability to incrementally learn new classes is vital to all real-world artificial intelligence systems. A large portion of high-impact applications like social media, recommendation systems, E-commerce platforms, etc. can be represented by graph models. In this paper, we investigate the challenging yet practical problem,Graph Few-shot Class-incremental (Graph FCL) problem, where the graph model is tasked to classify both newly encountered classes and previously learned classes. Towards that purpose, we put forward a Graph Pseudo Incremental Learning paradigm by sampling tasks recurrently from the base classes, so as to produce an arbitrary number of training episodes for our model to practice the incremental learning skill. Furthermore, we design a Hierarchical-Attention-based Graph Meta-learning framework, HAG-Meta from an optimization perspective. We present a task-sensitive regularizer calculated from task-level attention and node class prototypes to mitigate overfitting onto either novel or base classes. To employ the topological knowledge, we add a node-level attention module to adjust the prototype representation. Our model not only achieves greater stability of old knowledge consolidation, but also acquires advantageous adaptability to new knowledge with very limited data samples. Extensive experiments on three real-world datasets, including Amazon-clothing, Reddit, and DBLP, show that our framework demonstrates remarkable advantages in comparison with the baseline and other related state-of-the-art methods.

Original languageEnglish (US)
Title of host publicationWSDM 2022 - Proceedings of the 15th ACM International Conference on Web Search and Data Mining
PublisherAssociation for Computing Machinery, Inc
Pages987-996
Number of pages10
ISBN (Electronic)9781450391320
DOIs
StatePublished - Feb 11 2022
Event15th ACM International Conference on Web Search and Data Mining, WSDM 2022 - Virtual, Online, United States
Duration: Feb 21 2022Feb 25 2022

Publication series

NameWSDM 2022 - Proceedings of the 15th ACM International Conference on Web Search and Data Mining

Conference

Conference15th ACM International Conference on Web Search and Data Mining, WSDM 2022
Country/TerritoryUnited States
CityVirtual, Online
Period2/21/222/25/22

Keywords

  • Few-shot learning
  • Graph neural networks
  • Incremental learning

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Computer Science Applications
  • Software

Fingerprint

Dive into the research topics of 'Graph few-shot class-incremental learning'. Together they form a unique fingerprint.

Cite this