Timezone: »

 
Poster
Learning Symmetric Embeddings for Equivariant World Models
Jung Yeon Park · Ondrej Biza · Linfeng Zhao · Jan-Willem van de Meent · Robin Walters

Wed Jul 20 03:30 PM -- 05:30 PM (PDT) @ Hall E #534
in Poster Session 2 »

Incorporating symmetries can lead to highly data-efficient and generalizable models by defining equivalence classes of data samples related by transformations. However, characterizing how transformations act on input data is often difficult, limiting the applicability of equivariant models. We propose learning symmetric embedding networks (SENs) that encode an input space (e.g. images), where we do not know the effect of transformations (e.g. rotations), to a feature space that transforms in a known manner under these operations. This network can be trained end-to-end with an equivariant task network to learn an explicitly symmetric representation. We validate this approach in the context of equivariant transition models with 3 distinct forms of symmetry. Our experiments demonstrate that SENs facilitate the application of equivariant networks to data with complex symmetry representations. Moreover, doing so can yield improvements in accuracy and generalization relative to both fully-equivariant and non-equivariant baselines.

Keywords: DL: Graph Neural Networks · DL: Other Representation Learning · DL: Sequential Models, Time series

Author Information

Jung Yeon Park (Northeastern University)
Ondrej Biza (Northeastern University)
Linfeng Zhao (Northeastern University)
Jan-Willem van de Meent (University of Amsterdam)
Robin Walters (Northeastern University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors