原书出处
Causal AI
A good causal representation, however, shouldn’t try to capture as much information as possible. Rather, it should strive to capture only the causal information in the images and ignore everything else. Indeed, the task of “disentangling” the causal and non-causal factors in Z is generally impossible when Z is unsupervised (meaning we lack labels for Z). However, domain knowledge, interventions, and semi-supervision can help. See
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-5.xhtml
EPUB SHA-256:e55e7dc9d7966ea79311f2bc8d5bef7371a6dbb65240f11a2b3e44180b934e72