知识卡片

普通读书笔记卡 · 6595.d · 生成模型:GPT-6

软最大随机化依赖温度

理解等级:中级

有限行动的Boltzmann规则可写π(x)∝exp(αQ(x))。α=0时均匀,α增大时概率集中到最高价值行动;概率与指数价值而非价值本身成正比,需归一化及数值稳定处理。它提供随机化,但不自动保证有效长期探索。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Softmax rules A softmax decision rule randomly selects an action with a probability proportional to the resulting utility. Let’s define C (x ) as the probability of choosing the action x. Then C (x ) is defined as a probability value proportional to The noise parameter α modulates between the two extremes. When α=0, we have a uniform distribution on all the choices. As α

    参考来源

    - 位置:《Causal AI》第12章《因果决策与强化学习》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-12.xhtml) - 结论依据:摘录中关于软最大随机化依赖温度的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Softmax rules A softmax decision rule randomly selects an action with a probability proportional to the resulting utility. Let’s define C (x ) as the probability of choosing the action x. Then C (x ) is defined as a probability value proportional to The noise parameter α modulates between the two extremes. When α=0, we have a uniform distribution on all the choices. As α