paper-with-me

Papers

Batchless Normalization: How to Normalize Activations Across Instances with Minimal Memory Requirements

2022-12-30 · Benjamin Berger, Victor Uc Cetina

In training neural networks, batch normalization has many benefits, not all of them entirely understood. But it also has some drawbacks. Foremost is arguably memory consumption, as computing the batch statistics requires all instances within the batch to be processed simultaneously, whereas without batch normalization it would be possible to process them one by one while accumulating the weight gradients. Another drawback is that that distribution parameters (mean and standard deviation) are unlike all other model parameters in that they are not trained using gradient descent but require special treatment, complicating implementation. In this paper, I show a simple and straightforward way to address these issues. The idea, in short, is to add terms to the loss that, for each activation, cause the minimization of the negative log likelihood of a Gaussian distribution that is used to normalize the activation. Among other benefits, this will hopefully contribute to the democratization of AI research by means of lowering the hardware requirements for training larger models.

📄 PDF Abstract BibTeX arXiv:2212.14729

Code (1)

ichteltelch/Batchless 공식 구현 tf

Methods 이 논문이 사용한 방법론

Batch Normalization 설명 없음

Similar Papers 제목 키워드 기반

Normalizing the Normalizers: Comparing and Extending Network Normalization Schemes

2016-11-14 · Mengye Ren, Renjie Liao, Raquel Urtasun, Fabian H. Sinz 외

Normalization techniques have only recently begun to be exploited in supervised learning tasks. Batch normalization exploits mini-batch statistics to normalize the activations. This was shown to speed up training and res…

image-classificationImage ClassificationLanguage ModelingLanguage Modelling+1

Online Normalization for Training Neural Networks

2019-05-15 · NeurIPS 2019 12 · Vitaliy Chiley, Ilya Sharapov, Atli Kosson, Urs Koster 외

Online Normalization is a new technique for normalizing the hidden activations of a neural network. Like Batch Normalization, it normalizes the sample dimension. While Online Normalization does not use batches, it is as …

General Classificationimage-classificationImage ClassificationImage Segmentation+3

Positional Normalization

2019-07-09 · NeurIPS 2019 12 · Boyi Li, Felix Wu, Kilian Q. Weinberger, Serge Belongie

A popular method to reduce the training time of deep neural networks is to normalize activations at each layer. Although various normalization schemes have been proposed, they all follow a common theme: normalize across …

Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence

2021-06-07 · NeurIPS 2021 12 · Antoine Labatie, Dominic Masters, Zach Eaton-Rosen, Carlo Luschi

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 appeara…

Optimization Theory for ReLU Neural Networks Trained with Normalization Layers

2020-06-11 · ICML 2020 1 · Yonatan Dukler, Quanquan Gu, Guido Montúfar

The success of deep neural networks is in part due to the use of normalization layers. Normalization layers like Batch Normalization, Layer Normalization and Weight Normalization are ubiquitous in practice, as they impro…

Learning Theory