Measuring Orthogonality in Representations of Generative Models

Join us for our Lunch Talk in July 2025. We’re very happy that Robin Geyer from the Medical Informatics group will present his work on “Measuring Orthogonality in Representations of Generative Models".
The BIFOLD Lunch Talk series gives BIFOLD members and external partners the opportunity to engage in dialogue about their research in Machine Learning and Big Data. Each Lunch Talk offers BIFOLD members, fellows and colleagues from other research institutes the chance to present their research and to network with each other.
The Lunch Talk takes place at the TU Berlin. For further information on the Lunch Talks and registration, contact Dr. Laura Wollenweber via email.
Abstract: In unsupervised representation learning, models aim to distill essential features from high-dimensional data into lower-dimensional learned representations, guided by inductive biases. Understanding the characteristics that make a good representation remains a topic of ongoing research. Disentanglement of independent generative processes has long been credited with producing high-quality representations.
However, focusing solely on representations that adhere to the stringent requirements of most disentanglement metrics, may result in overlooking many high-quality representations, well suited for various downstream tasks. These metrics often demand that generative factors be encoded in distinct, single dimensions aligned with the canonical basis of the representation space.
Motivated by these observations, we propose two novel metrics: Importance-Weighted Orthogonality \IWO and Importance-Weighted Rank IWR. These metrics evaluate the mutual orthogonality and rank of generative factor subspaces. Throughout extensive experiments on common downstream tasks, over several benchmark datasets and models, IWO and IWR consistently show stronger correlations with downstream task performance than traditional disentanglement metrics. Our findings suggest that representation quality is closer related to the orthogonality of independent generative processes rather than their disentanglement, offering a new direction for evaluating and improving unsupervised learning models.