← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8982.a · 生成模型:GPT-6

散度方向决定所需信息

理解等级:中级

正向KL可以在不知真实数据密度数值时,用数据样本估计与参数有关的对数似然项;反向KL的期望由模型采样,并包含真实密度在这些样本处的值。只有数据样本时后者通常难以直接计算,交换参数顺序不是无害改写。

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

      KL(pdata ‖ pθ) is sometime called the forward KL divergence, and it measures the probability of the data under the model distribution. The reverse KL divergence, KL(pθ ‖ pdata), does the converse, measuring the probability of the model's samples under the data distribution; because we do not have access to the true data distribution, the reverse KL divergence usually cannot be computed.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第32章《生成模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c032.xhtml) - 结论依据:摘录中关于散度方向决定所需信息的定义、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:KL(pdata ‖ pθ) is sometime called the forward KL divergence, and it measures the probability of the data under the model distribution. The reverse KL divergence, KL(pθ ‖ pdata), does the converse, measuring the probability of the model's samples under the data distribution; because we do not have access to the true data distribution, the reverse KL divergence usually cannot be computed.