Tag: Continual Learning
All the articles with the tag "Continual Learning".
-
Less, but Better: Efficient Multilingual Expansion for LLMs via Layer-wise Mixture-of-Experts
本文提出LayerMoE算法,通过基于层间语言相似性的专家分配和路由分类器,实现了多语言LLM的高效扩展,以更少的参数显著提升新语言性能并减少旧语言遗忘。
-
Recurrent Knowledge Identification and Fusion for Language Model Continual Learning
本文提出Recurrent-KIF框架,通过内外循环机制动态估计参数重要性并迭代融合新旧知识,在持续学习中有效缓解灾难性遗忘并促进知识转移,实验验证其在多个大语言模型上的性能优势。
-
Scalable Strategies for Continual Learning with Replay
本文提出低秩适应(LoRA)、整合和顺序合并三种策略以提升持续学习的可扩展性,通过减少重放样本需求(最高65%)并结合高效微调技术,在图像分类任务中显著提高性能。
-
MoL for LLMs: Dual-Loss Optimization to Enhance Domain Expertise While Preserving General Capabilities
本文提出MoL框架,通过对领域语料使用CE损失和对通用语料使用KL散度损失的双重优化策略,显著提升大型语言模型的领域专长,同时有效保留通用能力,并在医学领域任务中取得优异表现。
-
TiC-LM: A Web-Scale Benchmark for Time-Continual LLM Pretraining
This paper introduces TiC-LM, a web-scale benchmark for time-continual LLM pretraining using 114 Common Crawl dumps, demonstrating that replay and autoregressive schedules can match Oracle retraining on general web data with less compute, though trade-offs persist across domains.