MiLorE-SSL: 在不遗忘的前提下扩展多语言能力的自监督模型
MiLorE-SSL: Scaling Multilingual Capabilities in Self-Supervised Models without Forgetting
AI总结：
MiLorE-SSL通过结合LoRA和软MoE机制，实现高效多语言自监督学习，有效缓解遗忘问题，仅用2.14%参数在新语言上取得优异表现。
摘要：
自监督学习（SSL）极大地推动了语音表示学习的发展，但多语言SSL模型仍受限于预训练过程中遇到的语言。从头开始重新训练以纳入新语言是计算上昂贵的，而连续训练时若没有缓解策略往往会导致灾难性遗忘。为此，我们提出了MiLorE-SSL，一个轻量级框架，结合LoRA模块与软混合专家（MoE）机制，实现高效的连续多语言训练。LoRA提供高效的低秩适应，而软MoE促进跨语言专家共享，减少跨语言干扰。为进一步缓解遗忘，我们引入有限的再学习数据，避免依赖大规模历史语料库。实验表明，MiLorE-SSL在新语言上表现强劲，并在现有语言上提升了模型能力，仅使用2.14%的可训练参数。
英文摘要：
Self-supervised learning (SSL) has greatly advanced speech representation learning, but multilingual SSL models remain constrained to languages encountered during pretraining. Retraining from scratch to incorporate new languages is computationally expensive, while sequential training without migitation strategies often leads to catastrophic forgetting. To address this, we propose MiLorE-SSL, a lightweight framework that combines LoRA modules with a soft mixture-of-experts (MoE) mechanism for efficient continual multilingual training. LoRA provides efficient low-rank adaptation, while soft MoE promotes flexible expert sharing across languages, reducing cross-lingual interference. To further mitigate forgetting, we introduce limited replay data from existing languages, avoiding reliance on large historical corpora. Experiments on ML-SUPERB demonstrate that MiLorE-SSL achieves strong performance in new languages and improves the ability in existing ones with only 2.14% trainable parameters.
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。
