Similarity of Processing Steps in Vision Model Representations

Matéo Mahaut, Marco Baroni

Recent literature suggests that the bigger the model, the more likely it is to converge to similar, ``universal'' representations, despite different training objectives, datasets,…