Papers › Sparse Sinkhorn Attention

Sparse Sinkhorn Attention

26 Feb 2020ICML 2020 1arXiv:2002.11296archive 2025-07-28

Yi Tay, Dara Bahri, Liu Yang, Donald Metzler, Da-Cheng Juan

We propose Sparse Sinkhorn Attention, a new efficient and sparse method for learning to attend. Our method is based on differentiable sorting of internal representations. Concretely, we introduce a meta sorting network that learns to generate latent permutations over sequences. Given sorted sequences, we are then able to compute quasi-global attention with only local windows, improving the memory efficiency of the attention module. To this end, we propose new algorithmic innovations such as Causal Sinkhorn Balancing and SortCut, a dynamic sequence truncation method for tailoring Sinkhorn Attention for encoding and/or decoding purposes. Via extensive experiments on algorithmic seq2seq sorting, language modeling, pixel-wise image generation, document classification and natural language inference, we demonstrate that our memory efficient Sinkhorn Attention method is competitive with vanilla attention and consistently outperforms recently proposed efficient Transformer models such as Sparse Transformers.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

lucidrains/sinkhorn-transformer mentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Document ClassificationImage GenerationLanguage ModelingLanguage ModellingNatural Language Inference

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Introduced by this paper: Sinkhorn Transformer, SortCut Sinkhorn Attention, Sparse Sinkhorn Attention

Absolute Position EncodingsAdamAttentionBPEDense ConnectionsDropoutFeedforward NetworkLSTMLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerReLUResidual ConnectionSeq2SeqSigmoid ActivationSinkhorn TransformerSoftmaxSortCut Sinkhorn AttentionSparse Sinkhorn AttentionTanh ActivationTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections