Skip to content
Hot eventLive

LINEUP:共享低秩因子替代每用户LoRA

1 reports1 sources16 hr ago updated

Get the story

AI overview

2026年10月5日,arXiv Machine Learning Theory(一手)报道LINEUP方法,提出将个性化能力从每用户LoRA中解耦:把可复用的低秩个性化因子与每用户适配状态分离,共享组件保持固定,目标用户只需优化8个标量。对照的私有LoRA配置使用419万每用户参数。该报道为目前唯一来源,尚无更早或后续报道可对比。

Generated from reports · updated 16 hr ago

Timeline

Follow the coverage from different angles.

Oct 5, 2026
  1. arXiv · Machine Learning Theory
    LINEUP:把个性化能力从每用户 LoRA 中解耦

    LINEUP 将可复用的低秩个性化因子与每用户适配状态解耦,目标用户只需优化 8 个标量,共享组件保持固定,而对照的私有 LoRA 配置使用 419 万每用户参数。

Heat trend

Current heat 7·Comparable peak 10(Oct 5)·Comparable change over 24 hours –

02.557.510Oct5Oct5Oct5Oct6

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.