Ghost Noise for Regularizing Deep Neural Networks
This work addresses the problem of improving regularization in deep neural networks for researchers and practitioners, offering a method to enhance generalization without train-test discrepancies, though it is incremental on existing GBN techniques.
The authors investigated the regularization effect of Ghost Batch Normalization (GBN) by analyzing its induced 'Ghost Noise' and proposed Ghost Noise Injection (GNI) as a new technique that mimics this noise without the drawbacks of small batch training. They experimentally showed that GNI provides greater generalization benefits than GBN, with improvements in settings like layer-normalized networks.
Batch Normalization (BN) is widely used to stabilize the optimization process and improve the test performance of deep neural networks. The regularization effect of BN depends on the batch size and explicitly using smaller batch sizes with Batch Normalization, a method known as Ghost Batch Normalization (GBN), has been found to improve generalization in many settings. We investigate the effectiveness of GBN by disentangling the induced ``Ghost Noise'' from normalization and quantitatively analyzing the distribution of noise as well as its impact on model performance. Inspired by our analysis, we propose a new regularization technique called Ghost Noise Injection (GNI) that imitates the noise in GBN without incurring the detrimental train-test discrepancy effects of small batch training. We experimentally show that GNI can provide a greater generalization benefit than GBN. Ghost Noise Injection can also be beneficial in otherwise non-noisy settings such as layer-normalized networks, providing additional evidence of the usefulness of Ghost Noise in Batch Normalization as a regularizer.