analiza aiqbench
This research introduces post-hoc repair methods for Graph Neural Networks by decoupling errors caused by poor mixture weights from suboptimal logit positioning. The study demonstrates that translating logits within a message-induced displacement set yields significant accuracy gains across diverse datasets and backbones, outperforming standard reweighting strategies. Ultimately, the findings suggest that relaxing strict evidence constraints in favor of direct translation offers a more effective path to correcting GNN predictions.
Local Evidence and Geometric Readout Repair in Trained GNNs
arXiv:2609.27092v1 Announce Type: cross Abstract: Many node-classification GNNs apply a linear classifier to a nonnegative mixture of local messages. An error can reflect either poor mixture weights or a reachable logit set poorly positioned for the classifier. We separate these causes with an exact-mass linear program and two learned post-hoc repairs. Every reweighted prediction has an equivalent centered logit translation, but only translations in a message-induced displacement set are realizable by reweighting. Across eight datasets, eight GNN backbones, and ten splits, mean accuracy rises from 62.6% for the frozen models to 63.8% with reweighting and 65.3% with set-conditioned translation. A parameter-matched node-only translator reaches 64.6%, showing that translation explains most of the gain while the message set supplies a smaller additional benefit. Although oracle reweighting can correct many errors, label-free reweighting captures little of this potential: local evidence is often present but hard to select, and relaxing the evidence constraint is more effective than learning within it.
Abstract:Many node-classification GNNs apply a linear classifier to a nonnegative mixture of local messages. An error can reflect either poor mixture weights or a reachable logit set poorly positioned for the classifier. We separate these causes with an exact-mass linear program and two learned post-hoc repairs. Every reweighted prediction has an equivalent centered logit translation, but only translations in a message-induced displacement set are realizable by reweighting. Across eight datasets, eight GNN backbones, and ten splits, mean accuracy rises from 62.6% for the frozen models to 63.8% with reweighting and 65.3% with set-conditioned translation. A parameter-matched node-only translator reaches 64.6%, showing that translation explains most of the gain while the message set supplies a smaller additional benefit. Although oracle reweighting can correct many errors, label-free reweighting captures little of this potential: local evidence is often present but hard to select, and relaxing the evidence constraint is more effective than learning within it.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2609.27092 [cs.LG]
(or arXiv:2609.27092v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.27092 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 249 cuvinte
Subiecte conexe
TraceCAD: Trace-Guided Repair for Agentic CAD Generation
arXiv · 5 aug. 2026
TRACE: Learning to Self-Calibrate Wireless Digital Twins from ISAC Measurements
arXiv · 29 sept. 2026
PEAfowl: Perception-Enhanced Multi-View Vision-Language-Action for Bimanual Manipulation
arXiv · 25 aug. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026