Learning Probabilistic Symmetrization for Architecture Agnostic Equivariance
Dec 10, 2023
Speakers
Jinwoo Kim
Speaker · 0 followers
Tien Dat Nguyen
Speaker · 0 followers
Ayhan Suleymanzade
Speaker · 0 followers
Hyeokjun An
Speaker · 0 followers
Seunghoon Hong
Speaker · 0 followers
About
We present a novel framework to overcome the limitations of equivariant architectures in learning functions with group symmetries. In contrary to equivariant architectures, the framework uses an arbitrary backbone (such as an MLP or a transformer) and symmetrizes it to be equivariant to given group by employing a small equivariant network that parameterizes the probabilistic distribution underlying the symmetrization. The distribution is end-to-end trained with the backbone which can maximize performance while reducing sample complexity of symmetrization. We show that this approach ensures not only equivariance to the given group but also universal approximation ability in expectation. We implement our method on a simple patch-based transformer backbone initialized from pretrained vision transformer, and test it for a wide range of symmetry groups including permutation and Euclidean groups and their combinations. Empirical tests show competitive results against tailored equivariant architectures, suggesting the potential for learning equivariant functions for diverse groups using a non-equivariant universal backbone. We further show evidence of enhanced learning in symmetric modalities, like graphs, when pretrained from non-symmetric modalities, like vision.We present a novel framework to overcome the limitations of equivariant architectures in learning functions with group symmetries. In contrary to equivariant architectures, the framework uses an arbitrary backbone (such as an MLP or a transformer) and symmetrizes it to be equivariant to given group by employing a small equivariant network that parameterizes the probabilistic distribution underlying the symmetrization. The distribution is end-to-end trained with the backbone which can maximize pe…
Organizer
NeurIPS 2023
Account · 648 followers
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Learning Transformer Programs
Watch later
Dan Friedman, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Wyze Rule: Federated Rule Dataset for Rule Recommendation Benchmarking
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Spectral Maps for Learning on Subgraphs
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Pre-Training Protein Encoder via Siamese Sequence-Structure Diffusion Trajectory Prediction
Watch later
Zuobai Zhang, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Minimax Risks and Optimal Procedures for Estimation under Functional Local Differential Privacy
Watch later
Lee Bonwoo, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Heavy-Tailed Self-Regularization in Deep Neural Networks
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%