TLDR AI · RESEARCH & INNOVATION

IMPROVED TEST-TIME ADAPTATION

arxiv.org24 min read

Summary from the TLDR AI newsletter

STAble Memory rePlay (STAMP) is a method that addresses distribution shifts between training and test data using only unlabeled test data. Unlike existing methods, STAMP excels in both recognizing known classes and rejecting outliers during inference.