## **Crafting Digital Twins: Computational Methods Across Environment, Health & Infrastructure**

:::info
There is a publicly available version of this note here: https://hackmd.io/ywR3EAujRQW60LihJAokhA

*That note :point_up: does not contain names.*
:::

### Value Proposition

The "Crafting Digital Twins: Computational Methods Across Environment, Health & Infrastructure" seminar series will be an interdisciplinary platform for TRIC-DT researchers to share and discuss about the computational methods, algorithms, and models that underpin Digital Twin technology in diverse fields. This series will provide a space for exchange of knowledge and methodologies across the themes of environment, health, and infrastructure, with the aim to cultivate a unified vocabulary and uncover synergies in the work of researchers across TRIC-DT.

By uniting researchers to discuss the methodological hurdles in their foundational work, this series will offer immediate value to each team member, enabling them to acquire practical insights applicable to their projects at any stage, no matter how "DT-ready" their individual projects are. Beyond the technical learnings, presenting to a broader peer group allows researchers to see common challenges from fresh perspectives, potentially igniting innovative solutions to their problems.

### Practical info
- Montly sessions (last X of the month)
- We start with one in november & december
- In December we hand over organisation of the next 4 months to one of the PDRAs
- 1h format with 2 presentations across different themes on shared topics. First presentations is 15 minutes and lays out the general methods question/challenge before going into theme-specific solution; 2nd presentation is 10 minutes.
- Presenters share their slides with each other beforehand
- Presenters share a "glossary" with jargon/terminology and definitions worth sharing to facilitate cross-domain communication
- 30 minutes are reserved for Q&A & discussion

### Session Topics

1. **Uncertainty quantification**: Allowing for more accurate and reliable predictions and simulations, which is crucial for making informed decisions.
- Gaussian mixture models (Environment: Ben Evans)
- Distributions through Emulators (Cristobal Rodero / Jose Lemus)
- Sensitivity Analysis

2. **Scaling-up**. How to create fast and efficient approximations of complex models, allowing for real-time simulations & predictions.
- Exploration of the role of emulators in bridging the gap between high-fidelity models and real-time requirements of digital twins. (Cristobal Rodero, Marina Strocchi)
- Autoemulate: Martin Stoffel
- Lack of labelled data (Environment: Louisa van Zeeland)
- (Physics based) reduced order models (Ludovica Cicci)

3. **Unsupervised and self-supervised learning:** These methods are vital for analysing and classifying unlabelled data, enabling the development of models that can adapt and learn from the environment, a key aspect of digital twins.
- E.g lack of labelled data how to go around it (Environment: Louisa van Zeeland)
- feature / object detection:
- Bayesian Guassian Mixture algorithm (Environment: Ben Evans)
- CNNs (Environment: Martin Rogers; Health: Abdul Qayyum)
- Class imbalances

4. **Generative models:** These are crucial for creating realistic simulations in digital twins, allowing for the modelling of various scenarios and conditions.
- Environment: Andrew McDonald
- cyclic GANs Health: Abdul Qayyum

5. **Data ingress and preprocessing:** Efficient data access, preprocessing, and real-time processing are foundational for the development and functioning of digital twins, ensuring that the models are fed with accurate and timely data.
- Preprocessing pipelines across themes?
- Environment: Oliver Strickson, James Byrne, Jonny Smith (bringing together multiple data sources for decision making)