`

Timezone: »

 
Boxhead: A Dataset for Learning Hierarchical Representations
Yukun Chen · Andrea Dittadi · Frederik Träuble · Stefan Bauer · Bernhard Schölkopf

Mon Dec 13 09:00 AM -- 10:00 AM (PST) @ None
Event URL: https://openreview.net/forum?id=OBoauWxefOP »

Disentanglement is hypothesized to be beneficial towards a number of downstream tasks. However, a common assumption in learning disentangled representations is that the data generative factors are statistically independent. As current methods are almost solely evaluated on toy datasets where this ideal assumption holds, we investigate their performance in hierarchical settings, a relevant feature of real-world data. In this work, we introduce \emph{Boxhead}, a dataset with hierarchically structured ground-truth generative factors. We use this novel dataset to evaluate the performance of state-of-the-art autoencoder-based disentanglement models and observe that hierarchical models generally outperform single-layer VAEs in terms of disentanglement of hierarchically arranged factors.

Author Information

Yukun Chen (Bloomberg)
Andrea Dittadi (Technical University of Denmark)
Frederik Träuble (Max Planck Institute for Intelligent Systems)
Stefan Bauer (Max Planck institute)
Bernhard Schölkopf (MPI for Intelligent Systems, Tübingen)

More from the Same Authors