ACL Findings2026
Model-Agnostic Meta Learning for Class Imbalance Adaptation
Abstract
Class imbalance is a widespread challenge in NLP tasks, significantly hindering robust performance across diverse domains and applications. We introduce Hardness-Aware Meta-Resample (HAMR) 1 , a unified framework that adaptively addresses both class imbalance and data difficulty. HAMR employs bi-level optimizations to dynamically estimate instancelevel weights that prioritize genuinely challenging samples and minority classes, while a neighborhood-aware resampling mechanism amplifies training focus on hard examples and their semantically similar neighbors. We validate HAMR on six imbalanced datasets covering multiple tasks and spanning biomedical, disaster response, and sentiment domains. Experimental results show that HAMR achieves substantial improvements for minority classes and consistently outperforms strong baselines. Extensive ablation studies demonstrate that our proposed modules synergistically contribute to performance gains and highlight HAMR as a flexible and generalizable approach for class imbalance adaptation.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference cites DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing
Builds on14
- LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu et al.ICLR 2022 · 18,833 citations
- Decoupling Representation and Classifier for Long-Tailed RecognitionBingyi Kang, Saining Xie, Marcus Rohrbach, Zhicheng Yan et al.ICLR 2020 · 1,496 citations
- Dice Loss for Data-imbalanced NLP TasksXiaoya Li, Xiaofei Sun, Yuxian Meng, Junjun Liang et al.ACL 2020 · 575 citations
- Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and InferenceBenjamin Warner, Antoine Chaffin, Benjamin Clavié, Orion Weller et al.ACL 2025 · 552 citations
- DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding SharingPengcheng He, Jianfeng Gao, Weizhu ChenICLR 2023 · 394 citations
Related papers
- A Re-Balancing Strategy for Class-Imbalanced Classification Based on Instance DifficultySihao Yu, Jiafeng Guo, Ruqing Zhang, Yixing Fan et al.CVPR 2022 · 42 citations
- Revive Re-weighting in Imbalanced Learning by Density Ratio EstimationJiaan Luo, Feng Hong, Jiangchao Yao, Bo Han et al.NeurIPS 2024 · 16 citations
- MESA: Boost Ensemble Imbalanced Learning with MEta-SAmplerZhining Liu, Pengfei Wei, Jing Jiang, Wei Cao et al.NeurIPS 2020 · 80 citations
- A Unified Loss for Handling Inter-Class and Intra-Class Imbalance in Medical Image SegmentationFeilong Xu, Feiyang Yang, Xiongfei Li, Xiaoli ZhangAAAI 2025 · 6 citations
- Difficulty-aware Balancing Margin Loss for Long-tailed RecognitionMinseok Son, Inyong Koo, Jinyoung Park, Changick KimAAAI 2025 · 9 citations
