知识卡片
对抗条件分布一致需要理想条件
若有足够容量、正确数据采样和理想优化,条件对抗目标可推动生成与真实条件分布一致。但有限判别器受骗不等于这些条件已满足,独立同分布抽样还依赖随机输入等采样机制。训练损失低不能直接证明模型已经给出真实条件样本。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 条件生成模型
原书出处 · 1 处
- · 查看原书依据
If both reasons are avoided, then it can be shown that the produced samples are iid with respect to the true conditional distribution of the data pdata(Y | x) (this follows from the analogous proof for unconditional GANs in [168], since that proof applies to modeling any arbitrary distribution over Y, including pdata(Y | x) for any x).
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:摘录中关于对抗条件分布一致需要理想条件的定义、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:If both reasons are avoided, then it can be shown that the produced samples are iid with respect to the true conditional distribution of the data pdata(Y | x) (this follows from the analogous proof for unconditional GANs in [168], since that proof applies to modeling any arbitrary distribution over Y, including pdata(Y | x) for any x).