\emph{DRIFT}: A Benchmark for Task-Free Continual Graph Learning with Continuous Distribution Shifts

2026-05-13

AI Summary

Existing continual graph learning methods rely on predefined task boundaries, rendering them ill-suited for real-world scenarios characterized by continuous distribution shifts and the absence of explicit task identifiers. This work proposes the first unified continual graph learning framework under a task-free setting, modeling the data stream as a time-evolving mixture of latent task distributions and capturing their dynamics through Gaussian parameterization. To facilitate systematic evaluation, we introduce DRIFT—a diverse benchmark encompassing dynamics ranging from abrupt task switches to smooth distributional drifts—and integrate representative continual learning approaches within this setting. Experimental results demonstrate a significant performance degradation of existing methods in the task-free scenario, revealing their implicit dependence on known task boundaries and thereby underscoring the necessity and effectiveness of both the proposed framework and benchmark.

Abstract

Continual graph learning (CGL) aims to learn from dynamically evolving graphs while mitigating catastrophic forgetting. Existing CGL approaches typically adopt a task-based formulation, where the data stream is partitioned into a sequence of discrete tasks with pre-defined boundaries. However, such assumptions rarely hold in real-world environments, where data distributions evolve continuously and task identity is often unavailable. To better reflect realistic non-stationary environments, we revisit continual graph learning from a task-free perspective. We propose a unified formulation that models the data stream as a time-varying mixture of latent task distributions, enabling continuous modeling of distribution drift. Based on this formulation, we construct DRIFT, a benchmark that spans a spectrum of transition dynamics ranging from hard task switches to smooth distributional drift through a Gaussian parameterization. We evaluate representative continual learning methods under this task-free setting and observe substantial performance degradation compared to traditional task-based protocols. Our findings indicate that many existing approaches implicitly rely on task boundary information and struggle under realistic task-free graph streams. This work highlights the importance of studying continual graph learning under realistic non-stationary conditions and provides a benchmark for future research in this direction. Our code is available at https://github.com/gqBond/DRIFT.

Problem

• continual graph learning
• task-free
• distribution shift
• non-stationary environments
• catastrophic forgetting

Innovation

• task-free continual learning
• graph neural networks
• distribution shift
• non-stationary environments
• continual graph learning