arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2211.04076cs.LGcs.AI
收藏

基于可训练前馈核的线性自注意力逼近

Linear Self-Attention Approximation via Trainable Feedforward Kernel

Czech Technical University in Prague(捷克布拉格技术大学)

Uladzislau Yorsh, Alexander Kovalenko

摘要:

为了追求更快的计算速度,高效的Transformers展示了令人印象深刻的各种方法-达到次二次注意力复杂度的模型可以利用稀疏性或输入的低秩近似来减少参与密钥的数量;其他降低复杂度的方法包括局部敏感散列,密钥池,额外的内存来存储压缩或与其他架构(如CNN)混合的信息。通常基于强大的数学基础,内核化方法允许线性复杂度近似注意力,同时保持高准确性。因此,在本文中,我们的目标是扩展可训练核方法的思想,以近似Transformer架构的自注意力机制。

英文摘要:

In pursuit of faster computation, Efficient Transformers demonstrate an impressive variety of approaches -- models attaining sub-quadratic attention complexity can utilize a notion of sparsity or a low-rank approximation of inputs to reduce the number of attended keys; other ways to reduce complexity include locality-sensitive hashing, key pooling, additional memory to store information in compacted or hybridization with other architectures, such as CNN. Often based on a strong mathematical basis, kernelized approaches allow for the approximation of attention with linear complexity while retaining high accuracy. Therefore, in the present paper, we aim to expand the idea of trainable kernel methods to approximate the self-attention mechanism of the Transformer architecture.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑