Compressible Dynamics in Deep Overparameterized Low-Rank Learning & Adaptation
深度过参数化低秩学习与适应中的可压缩动力学
机构 * Department of Electrical Engineering \& Computer Science, University of Michigan
AI总结 本研究提出Deep LoRA方法，通过深度过参数化低秩学习与适应中的可压缩动力学，提升语言模型微调效率并减少过拟合。
Comments Accepted at ICML'24 (Oral)
