Mode Normalization
This addresses a limitation in normalization methods for deep learning, particularly in multi-modal data scenarios, though it appears incremental as an extension of existing techniques.
The paper tackles the reduced effectiveness of batch normalization in multi-modal distributions by proposing a flexible normalization method that detects data modes on-the-fly and jointly normalizes samples with common features, demonstrating outperformance over BN and other techniques in single and multi-task datasets.
Normalization methods are a central building block in the deep learning toolbox. They accelerate and stabilize training, while decreasing the dependence on manually tuned learning rate schedules. When learning from multi-modal distributions, the effectiveness of batch normalization (BN), arguably the most prominent normalization method, is reduced. As a remedy, we propose a more flexible approach: by extending the normalization to more than a single mean and variance, we detect modes of data on-the-fly, jointly normalizing samples that share common features. We demonstrate that our method outperforms BN and other widely used normalization techniques in several experiments, including single and multi-task datasets.