CoVar: New Framework Improves Pseudo-Label Selection in Semi-Supervised Learning
Researchers have developed CoVar, a new method that selects reliable pseudo-labels in semi-supervised learning by combining confidence scores with variance measurements rather than relying on confidence alone. The approach addresses limitations of existing methods that struggle with model overconfidence and imbalanced datasets. This advancement could improve the performance of machine learning models trained on partially labeled data, which is common in real-world applications.
CoVar is a confidence-variance framework designed to improve pseudo-label selection in semi-supervised learning, a technique where models learn from both labeled and unlabeled data. Traditional methods rely primarily on maximum-confidence thresholds, but these can be unreliable when models are overconfident or when classes are imbalanced. The researchers derived a second-order cross-entropy approximation showing that reliable pseudo-labels occur when confidence is high and residual-class variance is low. The method embeds predictions into a two-dimensional space and uses spectral relaxation to separate reliable from unreliable predictions without requiring manually-tuned thresholds. Testing on multiple datasets including PASCAL VOC 2012, Cityscapes, and CIFAR variants showed clear improvements on segmentation tasks and competitive results on classification benchmarks, suggesting that variance information provides a useful signal complementary to confidence alone.
What's missing
The paper does not discuss computational overhead during training or provide detailed runtime comparisons with baseline methods. Additionally, the study does not explore how the method performs with extremely small labeled datasets or discuss potential failure modes when the variance signal itself becomes unreliable.
What different sources said
- arXiv cs.AICenter
CoVar: Confidence-Variance-Guided Pseudo-Label Selection for Semi-Supervised Learning
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.