Can I read Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence on EtoBox?

Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence by Labatie, Antoine; Masters, Dominic; Eaton-Rosen, Zach; Luschi, Carlo is a scholarly article available to read on EtoBox.

What is Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence about?

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.

Author

Labatie, Antoine; Masters, Dominic; Eaton-Rosen, Zach; Luschi, Carlo

Published

2021

Language

EN