Skip to content
Hot eventLive

Finsler Flow Matching 论文发布

1 reports1 sources3 hr ago updated

Get the story

AI overview

2026年10月9日,研究者在 arXiv(Artificial Intelligence 栏目)发布论文,提出 Finsler Flow Matching(FFM),用于从离散 Markov 转移图学习连续随机动力学,并做单快照轨迹推断。该方法用一阶与二阶局部矩构建受 Freidlin–Wentzell 作用启发的 Finsler 结构,再经 simulation-free score 与 flow matching 蒸馏为连续生成过程,采用动力学感知的测地线插值。这是该事件的最新报道,此前无其他报道,目前未见后续验证或独立评论。

Generated from reports · updated 3 hr ago

Timeline

Follow the coverage from different angles.

Oct 9, 2026
  1. arXiv · Artificial Intelligence
    Finsler Flow Matching:用动力学感知的测地线插值做单快照轨迹推断

    研究者提出 Finsler Flow Matching(FFM),从离散 Markov 转移图学习连续随机动力学。FFM 用一阶与二阶局部矩构建受 Freidlin–Wentzell 作用启发的 Finsler 结构,经 simulation-free score 与 flow matching 蒸馏为连续生成过程。

Heat trend

Current heat 9·Comparable peak 10(Oct 9)·Comparable change over 24 hours –

02.557.510Oct9Oct9Oct9Oct9

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.