Transformers are powerful sequence models, but require time and memory that
grows quadratically with the sequence length. In this paper we introduce sparse
factorizations of the attention matrix which reduce this to $O(n \sqrt{n})$. We
also introduce a) a variation on architecture and initialization to train
deeper networks, b) the recomputation of attention matrices to save memory, and
c) fast attention kernels for training. We call networks with these changes
Sparse Transformers, and show they can model sequences tens of thousands of
timesteps long using hundreds of layers. We use the same architecture to model
images, audio, and text from raw bytes, setting a new state of the art for
density modeling of Enwik8, CIFAR-10, and ImageNet-64. We generate
unconditional samples that demonstrate global coherence and great diversity,
and show it is possible in principle to use self-attention to model sequences
of length one million or more.

本文介绍了一种称为 Sparse Transformers 的神经网络架构，该架构通过稀疏的注意力矩阵因式分解和其他一些技术，可以更有效地处理长序列等任务，并在 Enwik8，CIFAR-10，和 ImageNet-64 等基准数据集上创造出新的最优表现。