Skip to content
Hot eventLive

FSPO:预算化LLM强化学习后训练的风险与帕累托控制

1 reports1 sources16 hr ago updated

Get the story

AI overview

2026年10月5日,arXiv(Artificial Intelligence,一手)发布FSPO研究。FSPO是一种面向预算化LLM强化学习后训练的反馈状态控制器,联合学习策略一致风险-to-go模型与长视界效用模型,以在给定资源预算下控制风险并维持帕累托可行。研究称,在匹配GRPO资源包络下,FSPO取得66.11%留出集准确率和59.43%分布外(OOD)准确率,优于最强自适应基线PB2的64.47%和57.03%。目前公开信息仅含该报道,未见更早或相互矛盾的数据。

Generated from reports · updated 16 hr ago

Timeline

Follow the coverage from different angles.

Oct 5, 2026
  1. arXiv · Artificial Intelligence
    FSPO:面向预算化 LLM RL 后训练的策略一致风险与 Pareto 可行控制

    FSPO 是一种面向预算化 LLM RL 后训练的反馈状态控制器,联合学习策略一致风险-to-go 模型与长视界效用模型。在匹配 GRPO 资源包络下,FSPO 取得 66.11% 留出集和 59.43% OOD 准确率,优于最强自适应基线 PB2 的 64.47% 和 57.03%。

Heat trend

Current heat 6·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.