# Breaking BERT: Evaluating and Optimizing Sparsified Attention

arXiv: https://arxiv.org/abs/2210.03841 · HTML: https://mlsanity.com/paper/2210.03841/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 40 references that are in the map.
- First posted on arXiv 7 Oct 2022.

## Builds on (most cited first, 27 of 27)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://mlsanity.com/paper/1804.07461/) (2018)
- [Big Bird: Transformers for Longer Sequences](https://mlsanity.com/paper/2007.14062/) (2020)
- [Reformer: The Efficient Transformer](https://mlsanity.com/paper/2001.04451/) (2020)
- [Generating Long Sequences with Sparse Transformers](https://mlsanity.com/paper/1904.10509/) (2019)
- [BERT Rediscovers the Classical NLP Pipeline](https://mlsanity.com/paper/1905.05950/) (2019)
- [What Does BERT Look At? An Analysis of BERT's Attention](https://mlsanity.com/paper/1906.04341/) (2019)
- [Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned](https://mlsanity.com/paper/1905.09418/) (2019)
- [Are Sixteen Heads Really Better than One?](https://mlsanity.com/paper/1905.10650/) (2019)
- [Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting](https://mlsanity.com/paper/1907.00235/) (2019)
- [Efficient Content-Based Sparse Attention with Routing Transformers](https://mlsanity.com/paper/2003.05997/) (2020)
- [A Deep Reinforced Model for Abstractive Summarization](https://mlsanity.com/paper/1705.04304/) (2017)
- [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://mlsanity.com/paper/2004.10643/) (2020)
- [Visualizing and Measuring the Geometry of BERT](https://mlsanity.com/paper/1906.02715/) (2019)
- [A Decomposable Attention Model for Natural Language Inference](https://mlsanity.com/paper/1606.01933/) (2016)
- [Long Short-Term Memory-Networks for Machine Reading](https://mlsanity.com/paper/1601.06733/) (2016)
- [Blockwise Self-Attention for Long Document Understanding](https://mlsanity.com/paper/1911.02972/) (2019)
- [Adaptively Sparse Transformers](https://mlsanity.com/paper/1909.00015/) (2019)
- [Adaptive Attention Span in Transformers](https://mlsanity.com/paper/1905.07799/) (2019)
- [Encoding Sentences with Graph Convolutional Networks for Semantic Role Labeling](https://mlsanity.com/paper/1703.04826/) (2017)
- [Star-Transformer](https://mlsanity.com/paper/1902.09113/) (2019)
- [Graph Convolutional Encoders for Syntax-aware Neural Machine Translation](https://mlsanity.com/paper/1704.04675/) (2017)
- [Combiner: Full Attention Transformer with Sparse Computation Cost](https://mlsanity.com/paper/2107.05768/) (2021)
- [Is Sparse Attention more Interpretable?](https://mlsanity.com/paper/2106.01087/) (2021)
- [Influence Patterns for Explaining Information Flow in BERT](https://mlsanity.com/paper/2011.00740/) (2020)
- [Top-down Tree Long Short-Term Memory Networks](https://mlsanity.com/paper/1511.00060/) (2015)

## Abstract (arXiv)

Transformers allow attention between all pairs of tokens, but there is reason to believe that most of these connections - and their quadratic time and memory - may not be necessary. But which ones? We evaluate the impact of sparsification patterns with a series of ablation experiments. First, we compare masks based on syntax, lexical similarity, and token position to random connections, and measure which patterns reduce performance the least. We find that on three common finetuning tasks even using attention that is at least 78% sparse can have little effect on performance if applied at later transformer layers, but that applying sparsity throughout the network reduces performance significantly. Second, we vary the degree of sparsity for three patterns supported by previous work, and find that connections to neighbouring tokens are the most significant. Finally, we treat sparsity as an optimizable parameter, and present an algorithm to learn degrees of neighboring connections that gives a fine-grained control over the accuracy-sparsity trade-off while approaching the performance of existing methods.