T3R: Deeper Test-Time Adaptation for Graph Neural Networks via Gradient Rotation
2026-06-29 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors address the problem that Graph Neural Networks (GNNs) often perform worse when facing new kinds of data because their settings are fixed after training. They propose a method called T3R that helps the GNN adjust itself during testing using unlabeled data, without needing expensive new labels. T3R uses special rotation matrices to better connect what the model learned before with the new task, allowing it to update more parts of the network. Their tests show improved accuracy on both regression and classification tasks compared to standard methods that do not adapt at test time.
Graph Neural NetworksTest-Time TrainingDistribution ShiftSelf-Supervised LearningRotograd MatricesSurrogate GradientsCross-Domain AdaptationMean Absolute ErrorFine-TuningRegression and Classification
Authors
Huy Truong, Alexander Lazovik, Victoria Degeler
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.