← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6603 · 生成模型:GPT-6

提议网络要用权重校正偏差

理解等级:高级

从q(K|P=p)抽K,再按结果机制抽Y时,重要性权重与p(K)p(P=p|K)/q(K|P=p)成比例,需提议支持覆盖目标且评分一致。学到的反向网络只是近似提议,不能省略权重并称精确后验样本。

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

      ratio of the probability of the sample under the conditioned model, over the probability of the sample under the proposal. Resampling the samples using these weights will generate samples from the target distribution. Pyro provides a utility for importance sampling, but because of the varying length of the generated sequences, it will be easier to implement importance sampling directly.

    参考来源

    - 位置:《Causal AI》第13章《因果与大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-13.xhtml) - 结论依据:摘录中关于提议网络要用权重校正偏差的定义、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:ratio of the probability of the sample under the conditioned model, over the probability of the sample under the proposal. Resampling the samples using these weights will generate samples from the target distribution. Pyro provides a utility for importance sampling, but because of the varying length of the generated sequences, it will be easier to implement importance sampling directly.