TL;DR本文介绍了一种应用于 ASR 任务的早期退出策略(EE),借此解决自我监督学习模型在大规模任务中存在的推理速度慢和网络反应过度的问题,并提出了两种新的策略来优化性能和速度之间的平衡。
Abstract
self-supervised learning (SSL) models reshaped our approach to speech,
language and vision. However their huge size and the opaque relations between
their layers and tasks result in slow inference and network overthinking, where
predictions made from the last layer of large models is w