Organizations: Baylor University Waco, Texas, USA · INRIA Rennes, Ile-de-Vilaine, France · University of Arkansas Fayetteville, Arkansas, USA · Meta Menlo Park, California, USA · NEC Laboratories America Princeton, New Jersey, USA
Backdoor attacks pose a persistent security threat to machine learning systems, yet their behavior under graph continual learning (GCL) remains underexplored. Conventional graph backdoor attacks often lose effectiveness as models are updated across successive tasks, creating the impression that continual updates naturally weaken malicious associations. This work reveals a contrasting vulnerability: knowledge-preservation mechanisms in GCL can also preserve early injected backdoors when the poisoned update is compatible with the clean-task signal being retained. For graph data, this risk is amplified by structural triggers, which are embedded into the message-passing paths of graph neural networks (GNNs) and can induce stable representation shifts across tasks. To expose this risk, we propose BadGCL, a model-agnostic framework for persistent backdoor attacks under a strict early-stage intervention setting. BadGCL learns topology-aware structural triggers using a gradient mutual-information objective that couples poisoned gradients with clean-task optimization signals, together with a propagation consistency objective that stabilizes trigger-induced effects through graph message passing. We provide theoretical analysis showing that gradient compatibility and propagation stability jointly support long-term backdoor persistence under benign continual updates. Extensive experiments across task-IL and class-IL protocols, four GCL benchmarks, and diverse victim systems show that BadGCL achieves persistent attack effectiveness with limited clean-utility degradation, substantially outperforming graph and continual-learning backdoor baselines and remaining robust against representative defenses.
Figures & tables
Figure 1. Heatmap of attack success rate (ASR, %) for eight graph backdoor poisoning methods on CoraFull dataset across 35 tasks, illustrating a progressive decay in attack effectiveness under continual learning setting.
Table 1. Overall final performance on the CoraFull and Reddit datasets under the task-IL protocol across three victim GCL systems (all values in %). w/o poisoning denotes the benign reference for each victim system. The benign reference is shown for comparison only and excluded from best/runner-up ranking; best and runner-up results are highlighted in bold and underline , respectively. ↑ indicates higher is better, while ↓ indicates lower is better.
Table 2. Overall final performance on the CoraFull , Reddit , and OGB-Arxiv datasets under the class-IL protocol across two victim GCL systems (all values in %).
Figure 2. Performance evolution ( AA and ASR across tasks) on Reddit dataset under the task-IL and class-IL protocols.
Task Sequence
Task-IL
Class-IL
AA (%) ↑
ASR (%) ↑
AA (%) ↑
ASR (%) ↑
Original
97.76
98.17
95.12
95.48
Reverse
97.72
98.36
94.96
95.15
Random (fixed first)
97.79 ± 0.04
98.14 ± 0.14
94.98 ± 0.08
95.56 ± 0.20
Full Random
97.73 ± 0.12
97.54 ± 0.80
94.74 ± 0.23
92.98 ± 0.53
Table 3. Task-sequence robustness on Reddit with PDGNNs.
Figure 3. ASR trajectories under task replacement at different continual stages for class-IL OGB-Arxiv dataset. The dashed lines denote the new trajectories after task replacement.
Table 4. Defense and ablation results on CoraFull with PDGNNs. (a) evaluates defenses against BadGCL; (b) tests structure regularization under defenses; (c) tests propagation ablation without defenses. Subscripts indicate the absolute change relative to the corresponding attacked model.
Figure 4. Ablation of the gradient-compatibility weight λMI on OGB-Products with PDGNNs. Points closer to the upper-right corner indicate better trade-offs between clean utility and persistent backdoor effectiveness. The star marks the selected weight, and the dashed vertical line indicates the reference AA of benign PDGNNs.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Table 5. Overall performance on the OGB-Arxiv and OGB-Products datasets under the task-IL protocol.
Table 6. Overall performance on the OGB-Products dataset under the class-IL protocol.
Figure 5. Accuracy matrices of BadGCL (left), P-Trojan (middle), and GCBA (right) on Reddit under (a) task-IL and (b) class-IL protocols.
Non-Exemplar Continual Graph Learning (NECGL) seeks to eliminate the privacy risks intrinsic to rehearsal-based paradigms by retaining solely class-level prototype representations rather than raw graph examples for mitigating catastrophic forgetting. However, this design choice inevitably precipitates feature drift. As a nascent alternative, Analytic Continual Learning (ACL) capitalizes on the intrinsic generalization properties of frozen pre-trained models to bolster continual learning performance. Nonetheless, a key drawback resides in the pronounced attenuation of model plasticity. To surmount these challenges, we propose Analytic Drift Resister (ADR), a novel and theoretically grounded NECGL framework. ADR exploits iterative backpropagation to break free from the frozen pre-trained constraint, adapting to evolving task graph distributions and fortifying model plasticity. Since parameter updates trigger feature drift, we further propose Hierarchical Analytic Merging (HAM), performing layer-wise merging of linear transformations in Graph Neural Networks (GNNs) via ridge regression, thereby ensuring absolute resistance to feature drift. On this basis, Analytic Classifier Reconstruction (ACR) enables theoretically zero-forgetting class-incremental learning. Empirical evaluation on four node classification benchmarks demonstrates that ADR maintains strong competitiveness against existing state-of-the-art methods.
Lei Song, Shihan Guan, Youyong Kong
School of Computer Science and Engineering, Southeast University, Nanjnig 211189, China
Continual learning (CL) enables models to acquire new knowledge from sequentially arriving tasks while retaining previously learned knowledge. However, in practical scenarios, task streams collected from untrusted sources may contain backdoor-poisoned samples, posing a critical challenge to the stability, plasticity, and security of continual learners. In this work, we investigate a challenging setting termed Continual Learning Under Backdoor Attack (CLUBA), where each incremental task may involve a small proportion of maliciously manipulated training samples. Unlike conventional continual learning or backdoor defense scenarios, CLUBA requires models to simultaneously mitigate catastrophic forgetting, preserve adaptation capability, and prevent the absorption of malicious supervision during sequential updates. To address this challenge, we propose a robust dynamic-expansion framework that integrates sample purification, selective recovery, and robust expert routing into a unified continual learning paradigm. Specifically, we introduce Bi-Prototype Purification (BPP) to identify suspicious samples by exploiting semantic discrepancies in feature space. Based on purified data, Gradient Discrepancy-based Robustness Optimization (GDBRO) selectively recovers informative poisoned samples through pseudo-label correction and gradient consistency evaluation, improving robustness while maintaining model plasticity. Furthermore, Robust Feature Consistency-based Expert Selection (RFCBES) constructs perturbation-aware class prototypes to enable reliable expert routing under corrupted or shifted inputs.
Keyu Lin, Fei Ye, Qihe Liu +2
University of Electronic Science and Technology of China, Chengdu 610054, China
Graph learning research has increasingly shifted toward continual graph learning (CGL), which better reflects real-world scenarios where graphs evolve over time. However, existing CGL methods largely assume clean supervision and overlook a critical challenge: the newly arriving portions of the graph are often noisy, due to annotation errors or adversarial corruption. This mismatch limits their applicability in practice. In this work, we study robust continual graph learning, where models must simultaneously handle catastrophic forgetting and noisy supervision in evolving graph data. We show that label noise introduces a new failure mode, catastrophic remembering, where models persistently reinforce corrupted knowledge across tasks. To address these challenges, we propose a Unified Flow-Oriented framework (UFO). First, UFO models conditional feature distributions via flow-based generative modeling and produces replay representations, mitigating forgetting without storing historical data. Second, UFO estimates instance-level reliability scores to distinguish clean from noisy nodes, reducing the impact of corrupted supervision and alleviating catastrophic remembering. Extensive experiments on four benchmark graph datasets under varying noise ratios demonstrate that UFO consistently outperforms existing methods in both accuracy and forgetting metrics. Code is available at: https://anonymous.4open.science/r/UFO.
Danhui Zhang, Zhe Wang, Qing Qing +6
Jilin University · Adelaide University · RMIT University +1