T3R: Deeper Test-Time Adaptation for Graph Neural Networks via Gradient Rotation

2026-06-29

AI Summary

Graph neural networks suffer significant performance degradation under distribution shifts, yet the absence of labels at test time hinders effective fine-tuning. To address this challenge, this work proposes a gradient rotation mechanism based on the Rotograd matrix that enhances affinity between the target task and self-supervised auxiliary tasks, thereby generating target-oriented proxy gradients. This enables deep test-time adaptation across nearly the entire model, overcoming the limitation of conventional test-time training methods that restrict updates to shallow layers. The proposed approach achieves a relative improvement of at least 9.37% on the OGB cross-domain classification benchmark and reduces mean absolute error by 0.172 on regression tasks, substantially outperforming existing methods.

Abstract

Graph Neural Networks (GNNs) deployed in real-world systems typically have fixed weights, often leading to degraded performance under distribution shifts. This issue can be mitigated by conventional fine-tuning, but in many real-world cases, collecting labeled data is expensive or infeasible. A potential approach is Test-Time Training (TTT), which adapts models' weights using unlabeled test data, yet it is typically limited to shallow updates that affect only a subset of model parameters. We propose T3R, leveraging multiple Rotograd matrices to improve task affinity between the target and auxiliary tasks, essential for effective test-time training. T3R further introduces a rotation technique that reorients self-supervised signals using these matrices to create surrogate gradients for the target task, allowing deeper adaptation across nearly the entire architecture. Empirically, T3R reduces MAE by 0.172 points over standard inference in regression datasets and achieves at least 9.37% relative improvement on cross-domain OGB classification benchmarks compared to models without adaptation. These results highlight the potential to develop an adaptation pipeline for graph-based systems, particularly in settings where conventional fine-tuning or retraining is infeasible.

Problem

• Graph Neural Networks
• Distribution Shift
• Test-Time Training
• Unlabeled Test Data
• Model Adaptation

Innovation

• Test-Time Adaptation
• Graph Neural Networks
• Gradient Rotation
• Self-Supervised Learning
• Rotograd