ReSCOPED

Pretrained models offer label-free out-of-distribution detection

Researchers have developed a method for detecting out-of-distribution inputs in machine learning models without requiring fine-tuning or labels. Their approach leverages the inherent geometric structure within frozen pr…