← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8997.b · 生成模型:GPT-6

摊销推断并非总是一次前向

理解等级:中级

直接条件网络把多次求解的成本摊到训练中,可快速从输入产生结果。自回归与扩散模型仍通常需要多步生成,CRF也可能有特定高效精确算法。快慢取决于模型和采样器,不能把所有条件生成都归为单次前向或精确推断的近似。

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

      The conditional random fields (CRFs) from chapter 29 are one kind of model that fits this definition. Now we will see some other ones. The big difference is that CRFs make predictions via thinking slow [244]: given a query observation, you run belief propagation or another iterative inference algorithm to arrive at a prediction. The conditional generative models we will see in this section think

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于摊销推断并非总是一次前向的定义、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:The conditional random fields (CRFs) from chapter 29 are one kind of model that fits this definition. Now we will see some other ones. The big difference is that CRFs make predictions via thinking slow [244]: given a query observation, you run belief propagation or another iterative inference algorithm to arrive at a prediction. The conditional generative models we will see in this section think