Lune

NeurIPS2024Top-tier venue

What Matters in Graph Class Incremental Learning? An Information Preservation Perspective

Jialu Li, Yu Wang, Pengfei Zhu, Wanyu Lin, Qinghua Hu

Abstract

Graph class incremental learning (GCIL) requires the model to classify emerging nodes of new classes while remembering old classes. Existing methods are designed to preserve effective information of old models or graph data to alleviate forgetting, but there is no clear theoretical understanding of what matters in information preservation. In this paper, we consider that present practice suffers from high semantic and structural shifts assessed by two devised shift metrics. We provide insights into information preservation in GCIL and find that maintaining graph information can preserve information of old models in theory to calibrate node semantic and graph structure shifts. We correspond graph information into low-frequency local-global information and high-frequency information in spatial domain. Based on the analysis, we propose a framework, Graph Spatial Information Preservation (GSIP). Specifically, for low-frequency information preservation, the old node representations obtained by inputting replayed nodes into the old model are aligned with the outputs of the node and its neighbors in the new model, and then old and new outputs are globally matched after pooling. For high-frequency information preservation, the new node representations are encouraged to imitate the near-neighbor pair similarity of old node representations. GSIP achieves a 10% increase in terms of the forgetting metric compared to prior methods on large-scale datasets. Our framework can also seamlessly integrate existing replay designs. The code is available through https://github.com/Jillian555/GSIP .

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

6 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Transferable Graph Condensation from the Causal Perspective cites Open Graph Benchmark: Datasets for Machine Learning on Graphs
  • Supermasks in Superposition cites Continual learning with hypernetworks
  • Federated Continual Graph Learning cites Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay
  • GCAL: Adapting Graph Models to Evolving Domain Shifts cites Open Graph Benchmark: Datasets for Machine Learning on Graphs
  • GCAL: Adapting Graph Models to Evolving Domain Shifts cites Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay
  • Graphs Help Graphs: Multi-Agent Graph Socialized Learning cites Open Graph Benchmark: Datasets for Machine Learning on Graphs
  • Graphs Help Graphs: Multi-Agent Graph Socialized Learning cites Federated Continual Graph Learning
  • Graphs Help Graphs: Multi-Agent Graph Socialized Learning cites Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers6

  1. Graphs Help Graphs: Multi-Agent Graph Socialized LearningJialu Li, Yu Wang, Pengfei Zhu, Wanyu Lin et al.NeurIPS 2025 · 2 citations
  2. Transferable Graph Condensation from the Causal PerspectiveHuaming Du, Yijie Huang, Su Yao, Yiying Wang et al.AAAI 2026
  3. Parameter-Masked Decoupled Optimization for Cross-Domain Class-Incremental LearningZiqi Gu, Chunyan Xu, Yangguang Liu, Wenxuan Fang et al.ICML 2026
  4. Federated Continual Graph LearningYinlin Zhu, Miao Hu, Di WuKDD 2025
  5. GCAL: Adapting Graph Models to Evolving Domain ShiftsZiyue Qiao, Qianyi Cai, Hao Dong, Jiawei Gu et al.ICML 2025

Builds on25

  1. Open Graph Benchmark: Datasets for Machine Learning on GraphsWeihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong et al.NeurIPS 2020 · 3,935 citations
  2. Continual learning with hypernetworksJohannes von Oswald, Christian Henning, João Sacramento, Benjamin F. GreweICLR 2020 · 412 citations
  3. Supermasks in SuperpositionMitchell Wortsman, Vivek Ramanujan, Rosanne Liu, Aniruddha Kembhavi et al.NeurIPS 2020 · 364 citations
  4. Hierarchical Decomposition of Prompt-Based Continual Learning: Rethinking Obscured Sub-optimalityLiyuan Wang, Jingyi Xie, Xingxing Zhang, Mingyi Huang et al.NeurIPS 2023 · 183 citations
  5. Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience ReplayFan Zhou, Chengtai CaoAAAI 2021 · 175 citations

Related papers

  1. Topology-aware Knowledge Preservation for Class-Incremental LearningHan Zang, Yongfeng Dong, Linhao Li, Liang Yang et al.AAAI 2026
  2. Replay-and-Forget-Free Graph Class-Incremental Learning: A Task Profiling and Prompting ApproachChaoxi Niu, Guansong Pang, Ling Chen, Bing LiuNeurIPS 2024 · 32 citations
  3. Subspace-Aware Feature Reshaping for Open-Set Graph Class-Incremental LearningWeichao Zhang, Shuai Zheng, Yeyu Yan, Zhizhe Liu et al.ICML 2026
  4. Towards Effective Open-set Graph Class-incremental LearningJiazhen Chen, Zheng Ma, Sichao Fu, Mingbin Feng et al.ACM MM 2025
  5. Streaming Graph Neural Networks with Generative ReplayJunshan Wang, Wenhao Zhu, Guojie Song, Liang WangKDD 2022 · 33 citations
Dusk over the sea between two cliffs drawn in fine vertical lines