# Fraternal Dropout

arXiv: https://arxiv.org/abs/1711.00066 · HTML: https://mlsanity.com/paper/1711.00066/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 31 Oct 2017.

## Built on by (most cited first, 14 of 14)

- [R-Drop: Regularized Dropout for Neural Networks](https://mlsanity.com/paper/2106.14448/) (2021)
- [Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders](https://mlsanity.com/paper/2411.01220/) (2024)
- [A Survey on Dropout Methods and Experimental Verification in Recommendation](https://mlsanity.com/paper/2204.02027/) (2022)
- [Preventing posterior collapse in variational autoencoders for text generation via decoder regularization](https://mlsanity.com/paper/2110.14945/) (2021)
- [Information Geometry of Dropout Training](https://mlsanity.com/paper/2206.10936/) (2022)
- [Generative Adversarial Training Can Improve Neural Language Models](https://mlsanity.com/paper/2211.09728/) (2022)
- [Dependency-based Mixture Language Models](https://mlsanity.com/paper/2203.10256/) (2022)
- [Augmentation-induced Consistency Regularization for Classification](https://mlsanity.com/paper/2205.12461/) (2022)
- [Cross-Model Comparative Loss for Enhancing Neuronal Utility in Language Understanding](https://mlsanity.com/paper/2301.03765/) (2023)
- [R-Block: Regularized Block of Dropout for convolutional networks](https://mlsanity.com/paper/2307.15150/) (2023)
- [An Ensemble Approach to Personalized Real Time Predictive Writing for Experts](https://mlsanity.com/paper/2308.13576/) (2023)
- [CR-SFP: Learning Consistent Representation for Soft Filter Pruning](https://mlsanity.com/paper/2312.11555/) (2023)
- [Invariance-powered Trustworthy Defense via Remove Then Restore](https://mlsanity.com/paper/2402.00304/) (2024)
- [Layer-wise Regularized Dropout for Neural Language Models](https://mlsanity.com/paper/2402.16361/) (2024)

## Abstract (arXiv)

Recurrent neural networks (RNNs) are important class of architectures among neural networks useful for language modeling and sequential prediction. However, optimizing RNNs is known to be harder compared to feed-forward neural networks. A number of techniques have been proposed in literature to address this problem. In this paper we propose a simple technique called fraternal dropout that takes advantage of dropout to achieve this goal. Specifically, we propose to train two identical copies of an RNN (that share parameters) with different dropout masks while minimizing the difference between their (pre-softmax) predictions. In this way our regularization encourages the representations of RNNs to be invariant to dropout mask, thus being robust. We show that our regularization term is upper bounded by the expectation-linear dropout objective which has been shown to address the gap due to the difference between the train and inference phases of dropout. We evaluate our model and achieve state-of-the-art results in sequence modeling tasks on two benchmark datasets - Penn Treebank and Wikitext-2. We also show that our approach leads to performance improvement by a significant margin in image captioning (Microsoft COCO) and semi-supervised (CIFAR-10) tasks.