核心概念
AdaProj introduces a novel loss function for learning class-specific subspaces, outperforming other methods in anomalous sound detection.
統計資料
AdaProjは他の一般的な損失関数を上回る性能を示す:DCASE2023データセットで73.6%のAUCと60.5%のpAUC。
引述
"AdaProj has arbitrari large optimal solution spaces allowing to relax the compactness requirements of the class-specific distributions."
"Using AdaProj results in better performance than other commonly used loss functions."