ETH Zurich

Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations

A large scale, comprehensive study challenges various assumptions in learning disentangled representations, which motivates demonstrating concrete benefits in robust experimental setups in future work.