Confusion-Aware Transfer Teacher Curriculum Learning Framework: Disentangling Scoring and Pacing Effects

2026-06-16

AI Summary

This work addresses the strong coupling between sample difficulty scoring and training schedule in curriculum learning, which hinders independent assessment of their individual contributions to model performance. Within the Transfer Teacher framework, the authors disentangle the effects of scoring mechanisms and training dynamics for the first time by leveraging phase-specific test subsets and a random-ordering baseline. They propose a confusion-aware difficulty scoring method that jointly considers the confidence of the correct class and the distribution of incorrect classes, aligning with human intuition and offering interpretability. Experiments demonstrate that the proposed method yields a sensible difficulty ranking on CIFAR-10 and, when using only 20% of the training data, achieves up to an 8.7% accuracy gain over random ordering, substantially improving data efficiency.

Abstract

Curriculum learning couples two design choices, how samples are scored by difficulty and how harder samples are paced into training, making it difficult to attribute observed gains to either component. We disentangle these factors with two evaluation protocols: stage-wise test subsets that validate scoring functions independently of curriculum training, and a baseline that applies the same pacing schedule to randomly ordered data. Within the Transfer Teacher framework (TTF), we use these protocols to evaluate a confusion-aware difficulty score that considers both correct-class confidence and the probability distribution over incorrect classes. On CIFAR-10 with ResNet-18 and VGG-16, the proposed score produces model-interpretable difficulty rankings that align with human intuition. However, at full data, neither curriculum nor anti-curriculum ordering improves accuracy over standard training, indicating that improving the scoring function alone is insufficient to overcome the known failure modes of curriculum learning in TTF. In contrast, We find that confusion-aware curriculum ordering result in consistent data-efficiency benefits, outperforming random ordering by up to 8.7% points at the 20% data regime, suggesting the potential of TTF as a data-efficient training method.

Problem

• curriculum learning
• difficulty scoring
• pacing
• disentanglement
• transfer teacher framework

Innovation

• curriculum learning
• confusion-aware scoring
• Transfer Teacher framework
• data efficiency
• difficulty disentanglement