Weakly Supervised Disentanglement with Guarantees
This work addresses the challenge of interpretable and human-controllable machine learning by offering a formalism to ensure disentanglement in weakly supervised settings, which is incremental as it builds on existing concerns about unsupervised methods.
The paper tackles the problem of learning disentangled representations with weak supervision by providing a theoretical framework to analyze when and how such supervision guarantees disentanglement, and empirically verifies these guarantees for methods like restricted labeling, match-pairing, and rank-pairing.
Learning disentangled representations that correspond to factors of variation in real-world data is critical to interpretable and human-controllable machine learning. Recently, concerns about the viability of learning disentangled representations in a purely unsupervised manner has spurred a shift toward the incorporation of weak supervision. However, there is currently no formalism that identifies when and how weak supervision will guarantee disentanglement. To address this issue, we provide a theoretical framework to assist in analyzing the disentanglement guarantees (or lack thereof) conferred by weak supervision when coupled with learning algorithms based on distribution matching. We empirically verify the guarantees and limitations of several weak supervision methods (restricted labeling, match-pairing, and rank-pairing), demonstrating the predictive power and usefulness of our theoretical framework.