A key problem in predicting aerosol formation is the enormous number of compounds, reactions and processes involved as well as the complexity of these compounds and the difficulty of measuring the processes. The VILMA project aims to understand the formation of aerosols by establishing a set of virtual instruments and models.
VILMA, or the Virtual Laboratory for Atmospheric Transformations at the Molecular Level, is a center of excellence funded by the Academy of Finland under the CoE program 2022-2029. The virtual lab offers researchers the opportunity to tackle many unsolved problems in atmospheric science, including identifying the reactions responsible for the formation and growth of organic nanoparticles.
VILMA combines three areas: theory and modeling of atmospheric phenomena, experiments, and machine learning and artificial intelligence.
– In the AI part of the project, our goal is to create digital twins of atmospheric processes combining physical simulators and a range of AI and machine learning models. Subsequently, we intend to study these processes with various AI-based tools. This is what we call a virtual laboratory, explains Associate Professor Kai Puolamäki of the University of Helsinki. Puolamäki also coordinates the Atmospheric AI program at FCAI.
One problem is that the physical simulations currently used are computationally heavy. One solution is to develop machine learning models that can quickly emulate slower physical simulations.
In other words, VILMA will pilot new research methods that take years to develop. There are no ready-made solutions, everything must be developed by researchers. Thanks to funding from the Academy, new ideas can now be tested and developed over the long term.
Puolamäki focuses on explainable AI, such as how digital twins work, or whether there are simpler but potentially more explainable models that could be used to describe relevant processes.
– It is also about knowing how to build these digital twins interactively, while understanding how they work. Managing uncertainty is another big issue. Can we trust the results, how to calculate the confidence intervals of the results, and is the model used outside its scope? These are surprisingly difficult problems, especially when dealing with complex machine learning models, says Puolamäki.
Serving sustainability goals and future generations of researchers
In terms of subject and objectives, the project complies with the sustainability objectives, even if sustainability is not directly studied in VILMA.
It is of course clear that this is basic research. The results of our work concern, for example, the modeling of air quality. However, we will find methods that can be applied in other fields as well as in research-based knowledge acquisition, says Puolamäki.
These new methods can have a significant impact on natural science and experimental research.
– We solve key problems in atmospheric science, for which new methods are needed. Again, from a computer science perspective, we are investigating how AI-based methods can be applied in the natural sciences, which can have a huge effect on the future conduct of such measurement-based research, says Puolamaki.
Ultimately, the objective is to acquire the ability to model the atmosphere and the world more broadly also outside of molecular transformations, the current objective of VILMA.
The aim is to disseminate AI results and methods, for example by publishing open access software and tools for the virtual laboratory. Doctoral and postdoctoral researchers are also recruited for the project.
The Center of Excellence also takes into account the general public. In the future, versions of the VILMA virtual laboratory suitable for science communication will also offer schoolchildren and the general public the opportunity to learn not only about atmospheric science, but also about the scientific method in general.
VILMA is a center of research excellence funded by the Academy of Finland and led by the University of Helsinki, University of Eastern Finland, University of Tampere and Aalto University.
This story originally appeared on the FCAI website.