arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.12746cs.CLeess.AS
收藏

Lamer-SSL:基于层意识的LoRA专家混合用于无需遗忘的持续多语言自监督模型扩展

Lamer-SSL: Layer-aware Mixture of LoRA Experts for Continual Multilingual Expansion of Self-supervised Models without Forgetting

The Chinese University of Hong Kong(香港中文大学)

Jing Xu, Minglin Wu, Xueyuan Chen, Xixin Wu, Helen Meng

AI总结:

Lamer-SSL通过层意识的LoRA专家混合与重放策略,实现持续多语言自监督模型扩展,保持性能并减少遗忘。

摘要:

尽管性能出色,自监督语音模型在泛化到新语言时往往遇到困难,并且在持续训练过程中容易遗忘之前学到的知识。为此,我们提出了Lamer-SSL,一种参数高效的框架,该框架集成了一个层意识的LoRA专家混合(Lamer)模块与重放策略。Lamer模块能够灵活地在共享和语言特定的表示之间进行平衡,而层意识的专家分配将更多专家分配给更深的层,这些层包含更丰富的语义信息。同时,重放策略通过最小的数据保留先前知识,缓解持续训练过程中的遗忘问题。在自动语音识别(ASR)和语言识别(LID)的实验中,证明Lamer-SSL能够有效地将自监督模型扩展到新语言,同时在仅可训练2.14%参数的情况下,保持对之前学习语言的强劲表现。

英文摘要:

Despite their impressive performance, self-supervised speech models often struggle to generalize to new languages and tend to forget previously acquired knowledge during continual training. To address this, we propose Lamer-SSL, a parameter-efficient framework that integrates a Layer-Aware MixturE of LoRA Experts (Lamer) module with a replay strategy. The Lamer module enables flexible balancing between shared and language-specific representations, while layer-aware expert allocation assigns more experts to deeper layers where semantic information is richer. Meanwhile, the replay strategy retains prior knowledge using minimal data, mitigating forgetting during continual training. Experiments on automatic speech recognition (ASR) and language identification (LID) demonstrate that Lamer-SSL extends self-supervised models to new languages effectively while maintaining strong performance on previously learned languages with only 2.14% parameters being trainable.

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。

↑