Regularization for Deep Learning: A Taxonomy

Jan Kukačka, Vladimir Golkov, Daniel Cremers

Abstract: Regularization is one of the crucial ingredients of deep learning, yet the term regularization has various definitions, and regularization methods are often studied separately from each other. In our work we present a novel, systematic, unifying taxonomy to categorize existing methods. We distinguish methods that affect data, network architectures, error terms, regularization terms, and optimization procedures. We identify the atomic building blocks of existing methods, and decouple the assumptions they enforce from the mathematical tools they rely on. We do not provide all details about the listed methods; instead, we present an overview of how the methods can be sorted into meaningful categories and sub-categories. This helps revealing links and fundamental similarities between them. Finally, we include practical recommendations both for users and for developers of new regularization methods.

TL;DR: Systematic categorization of regularization methods for deep learning, revealing their similarities.

Keywords: neural networks, deep learning, regularization, data augmentation, network architecture, loss function, dropout, residual learning, optimization

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:1710.10686/code)