DIB-OD: Preserving the Invariant Core for Robust Heterogeneous Graph Adaptation via Decoupled Information Bottleneck and Online Distillation

arXiv cs.LG / 4/14/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces DIB-OD, a new framework for robust adaptation of Graph Neural Networks (GNNs) across heterogeneous domains that suffer from severe distribution shifts.
DIB-OD explicitly decomposes learned representations into orthogonal invariant (task-relevant, domain-transcending) and redundant (domain-specific noise) subspaces to reduce negative transfer and catastrophic forgetting.
It uses an Information Bottleneck teacher–student distillation setup and the Hilbert–Schmidt Independence Criterion to isolate and preserve a stable invariant core during transfer.
A self-adaptive semantic regularizer is proposed to prevent core corruption in the target domain by dynamically gating the influence of labels based on predictive confidence.
Experiments on chemical, biological, and social network datasets show DIB-OD outperforming prior methods, especially on difficult inter-type domain transfers, with improved generalization and anti-forgetting behavior.

Abstract

Graph Neural Network pretraining is pivotal for leveraging unlabeled graph data. However, generalizing across heterogeneous domains remains a major challenge due to severe distribution shifts. Existing methods primarily focus on intra-domain patterns, failing to disentangle task-relevant invariant knowledge from domain-specific redundant noise, leading to negative transfer and catastrophic forgetting. To this end, we propose DIB-OD, a novel framework designed to preserve the invariant core for robust heterogeneous graph adaptation through a Decoupled Information Bottleneck and Online Distillation framework. Our core innovation is the explicit decomposition of representations into orthogonal invariant and redundant subspaces. By utilizing an Information Bottleneck teacher-student distillation mechanism and the Hilbert-Schmidt Independence Criterion, we isolate a stable invariant core that transcends domain boundaries. Furthermore, a self-adaptive semantic regularizer is introduced to protect this core from corruption during target-domain adaptation by dynamically gating label influence based on predictive confidence. Extensive experiments across chemical, biological, and social network domains demonstrate that DIB-OD significantly outperforms state-of-the-art methods, particularly in challenging inter-type domain transfers, showcasing superior generalization and anti-forgetting performance.