Data Science and autonomous experimentation
The increase in computational power and the growing amount of digitalized data are fostering a wide-spread application of machine learning algorithms, which allow a non-parametric approach to data statistics and pave the way to unprecedented applications of computer-based predictions.
![https://www.empa.ch/documents/55975/83322/small_monitoring_modeling_flat.jpg/7b9f221d-3528-b7b4-67a6-2b585e6a59a4?t=1724664693692&doAsUserId=mLYofz1ofIc4z2qdMCjTcg%3D%3D%3F_NewsPortlet_categoryId%3D56975](https://www.empa.ch/documents/55975/83322/small_monitoring_modeling_flat.jpg/7b9f221d-3528-b7b4-67a6-2b585e6a59a4?t=1724664693692&doAsUserId=mLYofz1ofIc4z2qdMCjTcg%3D%3D%3F_NewsPortlet_categoryId%3D56975)
As the name suggests, the use of innovative computer simulations pervades any aspect of the research at the Laboratory for Computational Engineering. This activity is strategic as we aim at positioning ourselves at the forefront of the research that integrates experiments, mechanistic modeling, and data-driven modeling to improve the reliability of predictions and the control of artificial-intelligence (AI) systems.
In addition to develop mechanistic models (which include, e.g., partial and stochastic differential equations, PDE and SDE; network models; compartment models; computational fluid dynamics, CFD; and atomistic simulations), we are committed to advance data-science methods for science and engineering and design next-generation AI and monitoring systems that combine mechanistic and data-driven models (physics-informed machine learning, PIML), and integrate sensor hardware and data analysis for autonomous experimentation (embodied machine learning).
We are particularly interested in application of deep-learning and reinforcement-learning algorithms to improve analysis of laboratory data and to design measurement systems that are capable of autonomously measuring, learning, and building reality models.
Among others, we are making use of the following methods:
- Physics-informed and multi-fidelity convolutional neural network (CNN)
- Reinforcement learning for optimization and autonomous experimentations
- TBW: “Reinforcement learning/autonomous systems”
- Principal component analysis (PCA) and proper orthogonal decomposition (POD)
- Symbolic regression
- Random forest (RF) classification and segmentation
- Model reduction
- Particle methods
- Droplet dispersion in airborne pathogen transmission
- Model Cascades for Stochastic Particle Simulations of Rarefied Polyatomic Gases