Constructing diverse model sources improves pseudo-label learning accuracy
Constructing Structured Decision Sources for Consensus-Based Pseudo-Label Learning
Machine Learning
Summary
Pseudo-label learning helps computers learn from unlabeled data by guessing the right answers, but only if different models provide varied viewpoints. The authors found that making multiple models that see the data in controlled different ways leads to better learning. They tested this approach on popular citation network datasets and showed it improves the quality of these guesses and sometimes boosts final accuracy. Their work shows the way models are built matters more than how many models you have in consensus learning.
What this means in practice
- •For graph machine learning engineers: Improve pseudo-label quality when training graph neural networks by constructing diverse decision sources with varied neighborhood structures.
- •For data scientists in network analysis: Enhance label predictions on citation or social network data by selecting complementary model sources to produce more reliable consensus labels.
Authors
Long Wang
Abstract
Consensus can make pseudo-label learning more reliable, but only when its predictors contribute genuinely different evidence. Multiple models that repeat the same boundary provide additional votes without additional information. We address this problem by con structing decision sources through controlled changes to within-class structure. Starting from a shared graph representation, we vary center granularity and neighborhood mixing, reproduce each resulting source to test its stability, and select a complementary subset using node pair coassignment. Unanimous predictions from the selected sources are then ranked for student training. On the public fixed splits of Cora, CiteSeer, and PubMed, evaluated with five random seeds, the constructed sources improve fixed-budget training pseudo-label precision by 1.19 to 4.39 percentage points over three conventionally initialized GCN sources. Under matched structural filters, three-source consensus is more precise than each constituent source in all 45 dataset slot eed comparisons. The gains are strongest in pseudo-label quality: downstream accuracy remains competitive but does not lead on every dataset. These results identify source construction rather than model count alone as an important design problem for consensus-based pseudo-label learning.