For building successful digital twins it is crucial to assimilate data from available sensors to estimate the state and/or parameters of the underlying system. Furthermore, such estimates should by accompanied by a quantification of the uncertainty, typically in shape of a posterior distribution. Due to the complexity of many systems (high-dimensional, nonlinear, non-Gaussian, etc.), one must resort to computationally expensive Bayesian sampling methods such as particle filters and Monte Carlo methods [1]. However, it is in general infeasible to run such approaches in real-time. In this talk, we present a non-intrusive methodology that speeds up Bayesian data assimilation by utilizing a particle filter approach and replacing the computational bottlenecks with deep learning-based surrogate models. The high-dimensionality is dealt with using Wasserstein autoencoders [2] with modified vision transformer layers to reduce the high-dimensional state to a low-dimensional latent state. Thereafter, a transformer is used for time stepping in the latent space. Embedding the surrogate models into a particle filter, enables data assimilation in the latent space resulting in a latent space posterior distribution. This ensures highly efficient real-time state and parameter estimation with uncertainty quantification [3]. The proposed methodology is showcased on two problems: leak localization in multi phase pipe flow with synthetic observations, and seabed identification based on water wave height data from a real-world experiment. REFERENCES [1] Evensen, Geir, Femke C. Vossepoel, and Peter Jan van Leeuwen. Data assimilation fundamentals: A unified formulation of the state and parameter estimation problem. Springer Nature, 2022. [2] Tolstikhin, Ilya, et al. ”Wasserstein auto-encoders.” arXiv preprint arXiv:1711.01558 (2017). [3] M ̈ucke, N.T. , Sanderse, B., Boht ́e, S., & Oosterlee, C. W. ”Markov Chain Generative Adversarial Neural Networks for Solving Bayesian Inverse Problems in Physics Applications.” arXiv preprint arXiv:2111.12408 (2021)