Timezone: »

Poster
Big Bird: Transformers for Longer Sequences
Manzil Zaheer · Guru Guruganesh · Kumar Avinava Dubey · Joshua Ainslie · Chris Alberti · Santiago Ontanon · Philip Pham · Anirudh Ravula · Qifan Wang · Li Yang · Amr Ahmed

Mon Dec 07 09:00 PM -- 11:00 PM (PST) @ Poster Session 0 #65
Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP. Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our theoretical analysis reveals some of the benefits of having $O(1)$ global tokens (such as CLS), that attend to the entire sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to 8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context, BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also propose novel applications to genomics data.

#### Author Information

##### Amr Ahmed (Google Research)

Amr Ahmed is a Senior Staff Research Scientist at Google. He received his M.Sc and PhD degrees from the School of Computer Science, Carnegie Mellon University in 2009 and 2011, respectively. He received the best paper award at KDD 2014 , the best Paper Award at WSDM 2014, the 2012 ACM SIGKDD Doctoral Dissertation Award, and a best paper award (runner-up) at WSDM 2012. He co-chaired the WWW'18 track on Web Content Analysis and served as an Area Chair for IJCAI 2019, SIGIR 2019, SIGIR 2018, ICML 2018, ICML 2017, KDD 2016, WSDM 2015, ICML 2014, and ICDM 2014. His research interests include large-scale machine learning, data/web mining, user modeling, personalization, social networks and content analysis.