Optimizing Class Distributions for Bias-Aware Multi-Class Learning
This work addresses bias and reliability issues in safety-critical multi-class learning scenarios, such as prioritizing certain classes, but it is incremental as it builds on existing training pipelines with minimal changes.
The paper tackles the problem of bias in multi-class image classification by proposing BiCDO, a framework that optimizes class distributions to enhance reliability and minimize bias, demonstrating improved balanced performance on datasets like CIFAR-10 and iNaturalist21 with models such as EfficientNet and ResNet.
We propose BiCDO (Bias-Controlled Class Distribution Optimizer), an iterative, data-centric framework that identifies Pareto optimized class distributions for multi-class image classification. BiCDO enables performance prioritization for specific classes, which is useful in safety-critical scenarios (e.g. prioritizing 'Human' over 'Dog'). Unlike uniform distributions, BiCDO determines the optimal number of images per class to enhance reliability and minimize bias and variance in the objective function. BiCDO can be incorporated into existing training pipelines with minimal code changes and supports any labelled multi-class dataset. We have validated BiCDO using EfficientNet, ResNet and ConvNeXt on CIFAR-10 and iNaturalist21 datasets, demonstrating improved, balanced model performance through optimized data distribution.