Im Kern eines Forschungskonsortiums arbeitetdas Höchstleistungsrechenzentrum der Uni Stuttgart (HLRS) an Simulationsmodellen,die zum Beispiel helfen könnten,die Ausbreitung einer gefährlichen Krankheit besser zu verstehen oder dasVerhalten von Flüchtlingsströmen vorherzusagen.
Somewhere in the world there is a breakout of some lethal viral disease that is spreading rapidly, threatening to become a pandemic. Health authorities around the globe need to take immediate action to contain the disease. But, which airports will they have to shut down?
And, where can specially-trained aid workers best be deployed? Of course, such decisions already have to be taken today. However, they are usually based on empirical data and a large portion of gut instinct, as the dissemination path of any given virus cannot be accurately predicted. The Center of Excellence for Global Systems Science (CoeGSS) in which the University of Stuttgart's High-Performance Computing Center plays a major role, is collaborating in a number of international projects that could pave the way to more reliable simulations of complex scenarios. However, Dr. Bastian Koller, Managing Director of the HLRS and Technical Project Coordinator for the CoeGSS, makes a placating hand gesture at this point. It will probably take quite some time before it actually becomes possible to accurately simulate the spread of a pandemic or the path of potential refugee flows in the wake of a war with a high degree of probability.
Data Quality is the Decisive Factor
Researchers from the international consortium, in which twelve partners from the German, Spanish, Italian, Swedish and Polish industrial and research sectors are involved, will primarily be working on other questions until the end of the project period in September 2018. The challenge, in this context, is not so much the necessary computing performance - after all, there is a super computer at the HLRS that can currently execute 7.4 billion calculation steps per second. In fact, the main thing the researchers are asking themselves is how best to deploy this power in the most expedient and effective manner. Koller explains the approach: “we're trying to use data from various sources to incorporate it into the simulation model. These include both official statistics and social media”.
In this context, the researchers first had to clarify what data they could utilize at all and which companies would, at least in an emergency situation, release their data temporarily. Apart from this, not all data proved to be equally suitable or reliable. According to Koller: “It is relatively clear that, in the case of Twitter, for example, some 20 to 30 per cent of all accounts are in fact held by so-called bots. These can easily skew the results”. Following the start of the project in October 2015, the CoeGSS researchers’ task, therefore, was to obtain an appraisal for the quality of the data and to identify parameters for a simulation,which are, of course, different for every problem.
A Question of Perspective
In addition to the Stuttgart-based IT specialists, the interdisciplinary team also includes social scientists, whose task was to model the scenarios, i.e., essentially to determine which data and parameters would be necessary to enable the computation of a given scenario. Also involved was Dialogik, a non-profit institute for communication and cooperation research in Stuttgart, which was founded by Professor Ortwin Renn, the former Scientific Director of the University of Stuttgart’s Research Center for Interdisciplinary Risk and Innovation Studies (ZIRIUS), whose task was to mediate between the various research groups. “The challenge was to unite the different approaches prevalent in the various disciplines such as IT and the social sciences”.
Yet, all of the participants soon realized that simply gathering vast amounts of data and feeding them into some arbitrary computational model on the super computer would not be an effective approach. “Problems cannot be solved through computational power alone”, as Koller emphasizes. On the contrary, in the worst-case scenario, the computer's performance will be offset. Koller provides an example: “We can always model a five-wheeled car, but, of course, that would make no sense”. Thus, deliberations always turn on the question of ramifications: “our models should reflect reality, which is why the quality of the data and the model are absolutely crucial”.
Just One Among Many Criteria
To find out what data a given computational model requires, and what properties it needs to have, the CoeGSS researchers concentrated on three simple scenarios that had already been relatively well researched, to model them into the simulation. For example, under the heading “Green Growth”, they studied the factors that motivate people to buy electric cars. If it were to prove possible to use the simulation to identify the city in which a major increase in electric cars could be expected then this would allow one to draw inferences about the necessary infrastructure, Koller explains.
In Koller's opinion, modelling actual developments on the basis of good-quality data, thereby determining which of the data and parameters are crucial, is the correct way to refine the models. The fact that the European Commission provided 4.5 million euro of funding for the project highlights the depth of interest there is in simulations of this kind. Thus, as Koller reports, government authorities have already indicated their interest in simulations of thebehavior of refugee flows, which will be developed in a follow-up project.
It will still take a few years until such simulations will be used in practice, and even then they will not form the only basis for decision- making but rather they will be considered as just one among several decision influencers for the resulting measures. “We’ll never achieve a hundred-per-cent probability, but we're working on achieving as high a probability as possible”, says Koller. As such, the value of simulations will increase, but, even in the long-term, the calculated results will remain just one basis for human decision making.