Skip to content

Translation in progress Some entries still include text in the original language.

Oct 12 · Monday

All papers

Found 1 paper
Filter3

System type

Models and APIs19AI Agent1coding agent0

Layer

Prompt layer4Model layer8Application layer220Training and data layer1
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 247papers have no tags yet and are not included in filtered results.
System type：AI AgentSystem type：coding agentLayer：Training and data layerClear all

Another 247papers do not have category tags yet and are not included in filtered results.

  1. Evaluation and benchmarksarXiv 2609.15315

    安全强化学习评测指标：SafeRLEval 开源评测套件

    提出 SafeRLEval，用越界率与代价偏差评测安全强化学习

    arXiv
    2609.15315
    Published
    Layer
    Training and data layer
    Setting
    Embodied systems and robotics
    Evaluated models
    PPO-Lag, P3O, FOCOPS
    AbstractSafeRLEval 用越界率、归一化偏差和安全等级补充 E[C]≤d，并要求同时看贪心部署。

    Spoor、Plaat 与 Moerland 认为，安全强化学习沿用 CMDP 的 E[C]≤d，只报告平均是否安全，看不出越界频率和幅度，也分不清训练期、带探索噪声的最终策略和贪心部署策略。他们提出评测框架 SafeRLEval：越界率 V、按安全界归一化的平均代价偏差 Dnorm、只在越界回合上计算的归一化幅度 Dnorm+。用 IQM 跨任务和安全界聚合。

    Paper analysisRead analysis
You have reached the end