Efficient Content-Based Sparse Attention with Routing Transformers

• Published in: Transactions of the Association for Computational Linguistics Vol. 9; pp. 53 - 68
• Main Authors: Roy, Aurko, Saffar, Mohammad, Vaswani, Ashish, Grangier, David
• Format: Journal Article
• Language: English
• Published: One Rogers Street, Cambridge, MA 02142-1209, USA MIT Press 01.02.2021; MIT Press Journals, The; The MIT Press
• Subjects: Attention; Complexity; Computation; Computational linguistics; Image processing; Language; Language modeling; Linguistics; Machine translation; Memory; Modelling; Neighborhoods; Neural networks; Source code; Sparsity; Transformers
• ISSN: 2307-387X; 2307-387X
• DOI: 10.1162/tacl_a_00353

Self-attention has recently been adopted for a wide range of sequence modeling problems. Despite its effectiveness, self-attention suffers from quadratic computation and memory requirements with respect to sequence length. Successful approaches to reduce this complexity focused on attending to local sliding windows or a small set of locations of content. Our work proposes to learn dynamic sparse attention patterns that avoid allocating computation and memory to attend to content unrelated to the query of interest. This work builds upon two lines of research: It combines the modeling flexibility of prior work on sparse attention with the efficiency gains from approaches based on sparse attention. Our model, the Routing Transformer, endows self-attention with a sparse routing module based on online -means while reducing the overall complexity of attention to ( ) from ( ) for sequence length and hidden dimension . We show that our model outperforms comparable sparse attention models on language modeling on (15.8 vs 18.3 perplexity), as well as on image generation on (3.43 vs 3.44 bits/dim) while using fewer self-attention layers. Additionally, we set a new state-of-the-art on the newly released data-set, obtaining a test perplexity of 33.2 with a 22 layer Routing Transformer model trained on sequences of length 8192. We open-source the code for Routing Transformer in Tensorflow.