期刊论文

Kehua Guo

英文

Pattern Recognition

0031-3203

2025

112448

10.1016/j.patcog.2025.112448

CRediT authorship contribution statement Zheng Wu: Writing – review & editing, Methodology, Writing – original draft, Data curation. Kehua Guo: Writing – review & editing, acquisition, Visualization, Conceptualization. Sheng Ren: Supervision, Data curation, Investigation, Conceptualization. Bin Hu: Project administration, Resources, Formal analysis. Xiangyuan Zhu: Writing – original draft, Software, Methodology, Validation, Resources. Rui Ding: Validation, Writing – original draft, Software.

本校为其他机构

Real-world data are typically long-tailed, causing neural networks to over-fit head classes and underperform on rare tails. We propose Dual-Level Balanced Learning (DBL), an efficient training framework that balances gradients at both the class and instance levels. DBL combines Class-aware Balancing (CB), which corrects class-level imbalance by re-weighting gradients according to prediction bias; Instance-aware Balancing (IB), which alleviates instance-level imbalance by emphasising the learning of hard examples; and a lightweight Cross-Level Collaboration (CC) scheme that harmonises the two l...