Miaowen Dong
PhD student
About me
As a member of Jan Gerken’s group, my research focuses on equivariant neural networks and data augmentation. I am broadly interested in understanding how symmetry can be exploited to design more efficient and reliable deep learning models.
With a background in applied mathematics and probabilistic methods, I aim to integrate these approaches to improve performance and explore new applications of deep learning models.
My research is supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP).
Education
MSc in Mathematics in Science and Engineering at Technical University of Munich (TUM), Germany (2025) Thesis: Multilevel Estimators for Rare Events with Selective Refinement Strategy Supervised by: Prof. Dr. rer. nat. Elisabeth Ullmann
Publications
Equivariance and Augmentation for Bayesian Neural Networks #
Miaowen Dong, Axel Flinth, Jan E. Gerken
Symmetries are important for many deep learning tasks, ranging from applications in the sciences to medical imaging. However, there is an ongoing debate about whether to impose symmetry constraints on the neural network architecture (yielding equivariant neural networks) or learn them from augmented training data. Although equivariant networks are well-studied theoretically, much less is known about data augmentation, since analyzing augmentation requires control over the training dynamics. Inspired by recent results that show that augmented infinite deep ensembles are exactly equivariant, we study data augmentation for Bayesian neural networks (BNNs) trained with variational inference. We focus on variational distributions in the exponential family and derive conditions under which exact equivariance is reached. We furthermore obtain bounds on the equivariance error and introduce three novel symmetrization techniques which boost the effect of data augmentation in this setting. We conduct extensive numerical experiments which show that one of our symmetrization methods (orbit expansion) outperforms the baseline in both equivariance and overall performance. Our code is available at https://github.com/dmw1998/augment-BNNs.
