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

TL;DR

D2ACE is a novel multi-label batch selection method that dynamically adapts to training progress and enhances label correlation modeling, leading to improved efficiency and accuracy in deep multi-label classification.

Contribution

It introduces dual dynamic metrics and adaptive correlation enhancement to address limitations of static and irrelevant label-based methods in multi-label batch selection.

Findings

D2ACE outperforms existing methods on tabular and image benchmarks.

It achieves better predictive performance and training efficiency.

The method effectively models label correlations with local context awareness.

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.