Skip to content
Hot storiesContinuously updated

国防科大与清华团队提出ICRPM模型

1 reports1 reporting sourcesUpdated 3 hours ago

Start here

AI overview

2026年11月6日,国防科技大学与清华大学团队提出ICRPM模型。该模型针对rollout预算优化,将近期任务与风险记录作为上下文,无需额外生成rollout即可直接预测策略更新后的下一步任务风险。在基于Qwen3-4B-Base的Countdown实验中,ICRPM在更难的CD4测试集上pass@1达到55.7%,两项平均准确率达65.4%,在各项表现上均优于GRESO与MoPPS。

AI-generated from reports · Updated 3 hours ago

Report timeline

Follow the reports to see different sides of the story.

Nov 6
  1. AITNT · AI头条
    rollout预算最该花在哪?国防科大与清华提出ICRPM前瞻预测任务风险

    国防科技大学与清华大学团队提出 ICRPM,将近期任务与风险记录作为上下文,无需额外生成 rollout 即可直接预测策略更新后的下一步任务风险。在基于 Qwen3-4B-Base 的 Countdown 实验中,ICRPM 在更难的 CD4 测试集上 pass@1 达到 55.7%,两项平均准确率达 65.4%,均优于 GRESO 与 MoPPS。

Story heat trend

Current heat 9·Comparable peak 10(Nov 6 17:00)·Comparable change, last 24 hours–

02.557.510Nov6Nov6Nov6Nov6

Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.