In representation learning, there has been recent interest in developing algorithms to disentangle the ground-truth generative factors behind data, and metrics to quantify how fully this occurs. However, these algorithms and metrics often assume that both representations and →