MetaBalance: High-Performance Neural Networks for Class-Imbalanced Data
This addresses a common issue in real-world applications like fraud detection and facial recognition, offering a novel solution to improve model performance on imbalanced datasets.
The paper tackles the problem of class-imbalanced data in neural networks, where standard methods lead to memorization of minority classes, and proposes MetaBalance, a meta-learning approach that uses outer-loop and inner-loop losses with different balancing strategies, achieving superior performance across tasks like image classification and fraud detection compared to popular re-sampling methods.
Class-imbalanced data, in which some classes contain far more samples than others, is ubiquitous in real-world applications. Standard techniques for handling class-imbalance usually work by training on a re-weighted loss or on re-balanced data. Unfortunately, training overparameterized neural networks on such objectives causes rapid memorization of minority class data. To avoid this trap, we harness meta-learning, which uses both an ''outer-loop'' and an ''inner-loop'' loss, each of which may be balanced using different strategies. We evaluate our method, MetaBalance, on image classification, credit-card fraud detection, loan default prediction, and facial recognition tasks with severely imbalanced data, and we find that MetaBalance outperforms a wide array of popular re-sampling strategies.