Kernel Normalized Convolutional Networks
This work addresses the problem of training convolutional neural networks effectively without batch normalization, which is crucial for scenarios with small batch sizes or differential privacy requirements, representing an incremental improvement over existing normalization methods.
The authors tackled the limitations of batch normalization in convolutional neural networks, such as poor performance with small batch sizes and incompatibility with differential privacy, by proposing kernel normalization and kernel normalized convolutional networks (KNConvNets). They demonstrated that KNConvNets achieve higher or competitive performance compared to batch normalization counterparts in image classification and semantic segmentation, and significantly outperform batch-independent competitors like layer and group normalization in non-private and differentially private training.
Existing convolutional neural network architectures frequently rely upon batch normalization (BatchNorm) to effectively train the model. BatchNorm, however, performs poorly with small batch sizes, and is inapplicable to differential privacy. To address these limitations, we propose the kernel normalization (KernelNorm) and kernel normalized convolutional layers, and incorporate them into kernel normalized convolutional networks (KNConvNets) as the main building blocks. We implement KNConvNets corresponding to the state-of-the-art ResNets while forgoing the BatchNorm layers. Through extensive experiments, we illustrate that KNConvNets achieve higher or competitive performance compared to the BatchNorm counterparts in image classification and semantic segmentation. They also significantly outperform their batch-independent competitors including those based on layer and group normalization in non-private and differentially private training. Given that, KernelNorm combines the batch-independence property of layer and group normalization with the performance advantage of BatchNorm.