The tendency of different neural networks to learn similar internal representations despite differences in architecture or training.