Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

negMIX: Negative Mixup for OOD Generalization in Open-Set Node Classification

Junwei Gong, Xiao Shen, Zhihao Chen, Shirui Pan, Xiao Wang, Xi Zhou

Abstract

Open-set node classification (OSNC) allows unlabeled test data to contain novel classes previously unseen in the labeled data. The goal is to classify in-distribution (ID) nodes into corresponding known classes and reject out-of-distribution (OOD) nodes as unknown class. Despite recent notable progress in OSNC, two challenges remain less explored, i.e., how to enhance generalization to OOD nodes, and promote intra-class compactness and inter-class separability. To tackle such challenges, we propose a novel Negative Mixup with Cross-Layer Graph Contrastive Learning (negMIX) model. Firstly, we devise a novel negative Mixup method purposefully crafted for the open-set scenario with theoretical justification, to enhance the model's generalization to OOD nodes and yield clearer ID/OOD boundary. Additionally, a unique cross-layer graph contrastive learning module is developed to maximize the prototypical mutual information between the same class nodes across different topological distance neighborhoods, thereby facilitating intra-class compactness and inter-class separability. Extensive experiments validate significant outperformance of the proposed negMIX over state-of-the-art methods in various scenarios and settings.

negMIX: Negative Mixup for OOD Generalization in Open-Set Node Classification

Abstract

Open-set node classification (OSNC) allows unlabeled test data to contain novel classes previously unseen in the labeled data. The goal is to classify in-distribution (ID) nodes into corresponding known classes and reject out-of-distribution (OOD) nodes as unknown class. Despite recent notable progress in OSNC, two challenges remain less explored, i.e., how to enhance generalization to OOD nodes, and promote intra-class compactness and inter-class separability. To tackle such challenges, we propose a novel Negative Mixup with Cross-Layer Graph Contrastive Learning (negMIX) model. Firstly, we devise a novel negative Mixup method purposefully crafted for the open-set scenario with theoretical justification, to enhance the model's generalization to OOD nodes and yield clearer ID/OOD boundary. Additionally, a unique cross-layer graph contrastive learning module is developed to maximize the prototypical mutual information between the same class nodes across different topological distance neighborhoods, thereby facilitating intra-class compactness and inter-class separability. Extensive experiments validate significant outperformance of the proposed negMIX over state-of-the-art methods in various scenarios and settings.
Paper Structure (20 sections, 29 equations, 9 figures, 5 tables, 1 algorithm)

This paper contains 20 sections, 29 equations, 9 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. The Proposed negMIX Model
  5. Negative Mixup for OOD Generalization
  6. Cross-layer Graph Contrastive Learning
  7. Experiments
  8. Open-set Node Classification Performance
  9. Analysis of negMIX
  10. Conclusion
  11. Parameter Sensitivity
  12. Algorithm of negMIX
  13. Details of Baselines
  14. Efficiency Analysis
  15. Proof of Theorem 1
  16. ...and 5 more sections

Figures (9)

  • Figure 1: Comparison between Conventional Positive Mixup and the proposed Negative Mixup. (a) Conventional Positive Mixup constructs a mixed-up OOD sample that lies between a labeled ID node and a potential OOD node, and assigns it with a soft positive label of both ID and OOD classes. (b) Our Negative Mixup constructs a mixed-up OOD sample close to OOD while far away from ID, and assigns it with a positive label of OOD class and a negative label of ID class.
  • Figure 2: Model architecture of negMIX which consists of negative Mixup and cross-layer GCL.
  • Figure 3: Illustration of Cross-layer GCL. Different colors correspond to different classes. Circles represent nodes and stars represent prototypes.
  • Figure 4: OSNC performance under (a) different training ratios and (b) different numbers of OOD classes.
  • Figure 5: Visualization of embeddings on AmazonPhoto. Grey color denotes the OOD class.
  • ...and 4 more figures