TaskNorm: Rethinking Batch Normalization for Meta-Learning
This addresses a critical bottleneck for researchers and practitioners in meta-learning, enabling fairer comparisons and better performance in image classification tasks.
The paper tackled the problem of batch normalization being ineffective in meta-learning due to its hierarchical nature, and developed TaskNorm, which consistently improved classification accuracy and reduced training time across fourteen datasets.
Modern meta-learning approaches for image classification rely on increasingly deep networks to achieve state-of-the-art performance, making batch normalization an essential component of meta-learning pipelines. However, the hierarchical nature of the meta-learning setting presents several challenges that can render conventional batch normalization ineffective, giving rise to the need to rethink normalization in this setting. We evaluate a range of approaches to batch normalization for meta-learning scenarios, and develop a novel approach that we call TaskNorm. Experiments on fourteen datasets demonstrate that the choice of batch normalization has a dramatic effect on both classification accuracy and training time for both gradient based and gradient-free meta-learning approaches. Importantly, TaskNorm is found to consistently improve performance. Finally, we provide a set of best practices for normalization that will allow fair comparison of meta-learning algorithms.