Outstanding paper at #NeurIPS22 main idea: is it possible to figure out when test data is coming from classes unknown during training? First they prove an impossibility theorem, then give positive/constructive results to characterize learnability of OOD. https://
openreview.net/pdf?id=sde_7Zz
GXOE
…

→ View original post on X — @reza_zadeh,