TLDR AI · RESEARCH & INNOVATION
IMPROVED TEST-TIME ADAPTATION
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.