#PCA

Zero-Training Upset: Filling the 256d Null Space with PCA Lifts EmbeddingGemma-2 Audio from Rank 26 to #1

A Reddit experiment shows how to lift EmbeddingGemma-2 from rank 26 to #1 on the MAEB audio benchmark without training: PCA-compress intermediate Conformer layers and inject into the 256d null space, leaving text embeddings unchanged while greatly improving audio clustering and emotion recognition.