D2ACE: Multi-Label Batch Selection Guided by Dual Dynamics and Adaptive Correlation Enhancement

2026-05-10

AI Summary

This work addresses the limitations of existing batch selection strategies in deep multi-label classification, which rely on static label weights and a single metric, thereby failing to capture the dynamic shifts in metric utility and label importance during training and neglecting local label dependencies. To overcome these issues, the authors propose D2ACE, a novel approach that employs staged Bernoulli mixture sampling to jointly optimize uncertainty and noise-robust difficulty while dynamically adjusting label weights across training epochs. Additionally, D2ACE incorporates a local context-aware correlation enhancement module to model instance-adaptive label dependencies. This method is the first to simultaneously capture dynamic characteristics at both the metric and label levels, achieving significant performance gains over current batch selection strategies on multiple image and tabular benchmarks, with improved prediction accuracy and more efficient modeling of label correlations.

Abstract

Batch selection is crucial for improving both training efficiency and predictive performance in deep multi-label classification (MLC). Existing batch selection methods typically rely on a single metric to assess instance importance and use static label weights to distinguish label significance, neglecting the dynamic evolution of metric utility and label significance during training. In addition, the method that explicitly exploits label correlations is largely affected by abundant irrelevant labels and insensitive to local label distributions. To address these issues, we propose D2ACE, a novel multi-label batch selection method guided by Dual Dynamics and Adaptive Correlation Enhancement. D2ACE explicitly captures metric and label-level training dynamics by combining stage-wise Bernoulli mixture sampling, which balances uncertainty and noise-resistant hardness, with dynamic label weighting to recalibrate label priorities at each epoch based on current metric statistics. Furthermore, D2ACE introduces a local context-aware correlation enhancement to focus on relevant labels with instance-adaptive dependencies. Extensive experiments on tabular and image benchmarks demonstrate that D2ACE outperforms existing batch selection approaches across various deep MLC models, achieving stronger predictive performance and more efficient correlation modeling.

Problem

• multi-label classification
• batch selection
• label correlation
• dynamic weighting
• training dynamics

Innovation

• Dual Dynamics
• Adaptive Correlation Enhancement
• Dynamic Label Weighting
• Batch Selection
• Multi-Label Classification