Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence
This addresses a key problem in deep learning for researchers and practitioners by providing a batch-independent alternative that mitigates expressivity issues without batch dependence.
The paper tackled the performance degradation of batch-independent normalization methods by identifying two failure modes in pre-activations and introduced Proxy Normalization to emulate batch normalization's behavior, achieving consistent matching or exceeding of its performance.
We investigate the reasons for the performance degradation incurred with batch-independent normalization. We find that the prototypical techniques of layer normalization and instance normalization both induce the appearance of failure modes in the neural network's pre-activations: (i) layer normalization induces a collapse towards channel-wise constant functions; (ii) instance normalization induces a lack of variability in instance statistics, symptomatic of an alteration of the expressivity. To alleviate failure mode (i) without aggravating failure mode (ii), we introduce the technique "Proxy Normalization" that normalizes post-activations using a proxy distribution. When combined with layer normalization or group normalization, this batch-independent normalization emulates batch normalization's behavior and consistently matches or exceeds its performance.