Batch Renormalization: Towards Reducing Minibatch Dependence in Batch-Normalized Models
This addresses a key limitation in deep learning training for practitioners dealing with small batch sizes or non-i.i.d. data, offering an incremental improvement over existing methods.
The paper tackles the problem of Batch Normalization's reduced effectiveness with small or non-i.i.d. minibatches by proposing Batch Renormalization, which ensures training and inference outputs depend on individual examples, resulting in substantially better performance in such scenarios.
Batch Normalization is quite effective at accelerating and improving the training of deep models. However, its effectiveness diminishes when the training minibatches are small, or do not consist of independent samples. We hypothesize that this is due to the dependence of model layer inputs on all the examples in the minibatch, and different activations being produced between training and inference. We propose Batch Renormalization, a simple and effective extension to ensure that the training and inference models generate the same outputs that depend on individual examples rather than the entire minibatch. Models trained with Batch Renormalization perform substantially better than batchnorm when training with small or non-i.i.d. minibatches. At the same time, Batch Renormalization retains the benefits of batchnorm such as insensitivity to initialization and training efficiency.