arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.02596cs.LGcs.CLcs.SDeess.AS
收藏

基于BEST-RQ的线性复杂度注意力替代方案分析

An Analysis of Linear Complexity Attention Substitutes with BEST-RQ

Laboratoire Informatique d'Avignon(阿维尼翁信息学实验室);Avignon Université(阿维尼翁大学);Samsung AI Center Cambridge(三星剑桥人工智能中心);Inria(法国国家信息与自动化研究所);CNRS(法国国家科学研究中心);Grenoble INP(格勒诺布尔工业大学);LIG(信息与信号处理实验室)

Ryan Whetten, Titouan Parcollet, Adel Moumen, Marco Dinarelli, Yannick Estève

AI总结:

本文研究在BEST-RQ自监督语音学习中用HyperMixing、Fastformer、SummaryMixing和Mamba等线性复杂度方案替代多头自注意力,发现其能在保持性能的同时显著降低显存并提升速度。

摘要:

自监督学习(SSL)已被证明在包括语音处理在内的多个领域中有效。然而,SSL在计算和内存方面开销较大,部分原因在于多头自注意力(MHSA)的二次复杂度。针对MHSA的替代方案已被提出并应用于语音领域,但在SSL环境下尚未得到充分研究。在本工作中,我们研究了将MHSA替换为近期最先进的线性复杂度替代方案的效果,具体包括HyperMixing、Fastformer、SummaryMixing和Mamba。我们通过考察速度、显存(VRAM)消耗量以及在SSL MP3S基准上的性能来评估这些方法。结果表明,与MHSA相比,这些线性替代方案保持了有竞争力的性能,同时在输入序列长度为20到80秒的情况下,平均将VRAM消耗降低了约20%至60%,并将速度提升了7%至65%。

英文摘要:

Self-Supervised Learning (SSL) has proven to be effective in various domains, including speech processing. However, SSL is computationally and memory expensive. This is in part due the quadratic complexity of multi-head self-attention (MHSA). Alternatives for MHSA have been proposed and used in the speech domain, but have yet to be investigated properly in an SSL setting. In this work, we study the effects of replacing MHSA with recent state-of-the-art alternatives that have linear complexity, namely, HyperMixing, Fastformer, SummaryMixing, and Mamba. We evaluate these methods by looking at the speed, the amount of VRAM consumed, and the performance on the SSL MP3S benchmark. Results show that these linear alternatives maintain competitive performance compared to MHSA while, on average, decreasing VRAM consumption by around 20% to 60% and increasing speed from 7% to 65% for input sequences ranging from 20 to 80 seconds.

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。

↑