Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification

2026-05-22 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

Transductive Sharpening (TS) is a novel loss-level modification for semi-supervised node classification, designed to exploit predictions on unlabeled nodes in a transductive setting. It minimizes prediction entropy on unlabeled nodes while counterbalancing this effect on labeled nodes, using Tsallis entropy (Gini impurity) for stable optimization. Evaluated across 13 node-classification benchmarks, TS consistently improves performance for Graph Neural Networks (GCN, GAT, GraphSAGE) and Multi-Layer Perceptrons (MLPs), adding only a single scalar hyperparameter, lambda. A universal lambda of 0.25 proved effective, demonstrating robustness without architecture changes or significant computational overhead.

Key takeaway

For Machine Learning Engineers developing node classification models in transductive settings, Transductive Sharpening offers a simple, architecture-agnostic method to boost performance. By exploiting unlabeled node predictions through a loss-level modification, you can achieve consistent gains across various GNNs and MLPs. Consider integrating TS with a conservative lambda, such as 0.25, to enhance model confidence and accuracy without increasing architectural complexity or computational overhead.

Key insights

Discarded unlabeled node predictions provide a valuable signal for improving transductive graph learning.

Principles

• Unlabeled predictions offer a direct training signal.
• Symmetric uncertainty shaping prevents overconfidence.
• Moderate sharpening improves decision boundaries.

Method

TS augments supervised loss by adding a term that minimizes Tsallis entropy on unlabeled nodes and maximizes it on labeled nodes, controlled by a scalar hyperparameter lambda.

In practice

• Apply TS to existing GNN or MLP architectures.
• Set the sharpening coefficient lambda between 0 and 0.5.
• Use Tsallis entropy (Gini impurity) for stable training.

Topics

• Transductive Node Classification
• Graph Neural Networks
• Loss Functions
• Entropy Minimization
• Semi-supervised Learning
• Tsallis Entropy

Code references

• transductive-sharpening/tunedGNN

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Semi-Supervised Learning on Graphs using Graph Neural Networks — GNN performance in semi-supervised learning is bounded by approximation, stochastic, and optimization errors, influenced by label scarcity and graph topology.
• Multimodal Graph Negative Learning — Multimodal Graph Negative Learning guides models by teaching unlikely classes, avoiding bias propagation from unreliable dominant branches.
• Scaling Learning-based AEB with Massive Unlabeled Data — Stabilized meta-feedback semi-supervised learning effectively scales AEB using massive unlabeled data, significantly improving safety metrics in real-world deployment.
• The Label Imitation Game: Turing Test Network for Zero-Shot Pseudo-Label Pruning — The Turing Test Network (TTN) prunes zero-shot pseudo-labels by evaluating them contextually, improving accuracy and enabling zero-shot task transfer.
• A Zero-shot Generalized Graph Anomaly Detection Framework via Node Reconstruction — AlignGAD enables zero-shot graph anomaly detection across diverse domains by unifying features and leveraging reconstruction discrepancy.
• Symmetrization of Loss Functions for Robust Training of Neural Networks in the Presence of Noisy Labels — Symmetrizing multi-class loss functions, especially cross-entropy, yields a unique convex unhinged loss robust to noisy labels.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.