知识卡片
噪声后验须与潜变量样本配对
若解码概率p依赖潜变量z,抽取像素噪声的条件区间就应与同一个后验z及事实图像对应。用z均值算区间、再独立抽另一个z生成,会打破这一条件关系;不能只保存彼此独立的边缘参数来代表完整溯因。
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
原书出处 · 1 处
- · 查看原书依据
n_unif_upper = decoder.forward( #3 n_image_loc, #3 encode_one_hot(factor) #3 ) #3 n_unif_upper = n_unif_upper * (1 - 2 * smoother) + smoother #3 p_image_params = n_image_loc, n_image_scale, n_unif_upper #4 return p_image_params def do_action(factor, element=1, val=2): #5 intervened_factor = factor.clone() #5 intervened_factor[0][element] = val #5 return intervened_factor #5 def predict(intervened_factor, n_image_params): #6 n_image = p_n_image(n_image_params) #6 sim_img = f_image(intervened_factor, n_image) #6 return sim_img #6 def counterfactual(img, factor):
参考来源
- 位置:《Causal AI》第09章《通用反事实推断算法》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-9.xhtml)
- 结论依据:摘录中关于噪声后验须与潜变量样本配对的定义、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:n_unif_upper = decoder.forward( #3 n_image_loc, #3 encode_one_hot(factor) #3 ) #3 n_unif_upper = n_unif_upper * (1 - 2 * smoother) + smoother #3 p_image_params = n_image_loc, n_image_scale, n_unif_upper #4 return p_image_params def do_action(factor, element=1, val=2): #5 intervened_factor = factor.clone() #5 intervened_factor[0][element] = val #5 return intervened_factor #5 def predict(intervened_factor, n_image_params): #6 n_image = p_n_image(n_image_params) #6 sim_img = f_image(intervened_factor, n_image) #6 return sim_img #6 def counterfactual(img, factor):