Digital twins are virtual representations of physical assets, created using parameterized computational models that are specifically tailored to mirror their real-world counterparts. To accurately predict the health and behaviour of the physical asset over time, observational data must be leveraged to update the virtual representation. Formally, this amounts to performing data assimilation, which often involves solving an inverse problem. While inverse problems for large-scale complex systems have been the focus of much work [1-2], digital twins present new challenges and opportunities. In particular, enabling digital twins requires methodologies that: (1) leverage multi-modal data or data across fleets of assets, (2) estimate both reducible (epistemic) and irreducible (aleatoric) uncertainties, (3) account for down-stream decision making goals on actionable time scales, and (4) enable real-time updates to digital twin systems. This minisymposium welcomes presentations on recent research addressing such challenges. Topics of interest include (but are not limited to) data-assimilation, data-driven approaches, frequentist or Bayesian formulations, ensemble-based approaches, and optimal experimental design.