Search papers, labs, and topics across Lattice.
The paper introduces Hardness-Aware Meta-Resample (HAMR), a meta-learning framework to address class imbalance by dynamically weighting samples based on their difficulty and class representation. HAMR uses bi-level optimization to estimate instance weights, focusing on challenging samples and minority classes, coupled with neighborhood-aware resampling to amplify training on hard examples and similar neighbors. Experiments across six imbalanced NLP datasets show HAMR substantially improves minority class performance and outperforms strong baselines, demonstrating its generalizability.
Forget re-sampling heuristics – HAMR uses meta-learning to dynamically re-weight training examples based on difficulty and class representation, leading to significant gains on imbalanced NLP tasks.
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), a unified framework that adaptively addresses both class imbalance and data difficulty. HAMR employs bi-level optimizations to dynamically estimate instance-level 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. Code is available at https://github.com/trust-nlp/ImbalanceLearning.
