Where should you plant ten trees so that as many city dwellers as possible can enjoy them? If a smart algorithm knows how people move through the city AND where there are already trees, it can calculate the optimal solution. Data scientist Mitra Baratchi makes it possible. Her students are now studying corona policy using this method.
Author: Rianne Lindhout
Much data is simply public; anyone can go to the online satellite data portal, for example, or to data.overheid.nl. Huge amounts of data are also available via smartphones: wifi usage tells you, for example, where a festival site is very crowded. We look at that with suspicion, but you can also do a lot of good with it. Make optimal policy on solutions to traffic congestion, for example.
Mitra Baratchi is figuring out how to extract useful information from data. In a variety of fields, she is pulling the right data together into self-designed algorithms that help computer programs learn. Those programs can then propose automatically generated solutions.
Within the Automated Design of Algorithms (ADA, part of LIACS) research group, Baratchi focuses on spatio-temporal data: data about space and time. She set up the Urban computing course for master's students to open their eyes to the possibilities. "Suppose an entrepreneur wants to open a store in a city. He wants it preferably in a place where as many people as possible pass by during the day. The city government, on the other hand, wants to spread out the crowds a bit and also doesn't want all the stores to be in a clump.' In the international NetMob Future cities challenge, Baratchi's students designed an algorithm that uses data on store locations and mobility data to make proposals on optimal locations. Locations that entrepreneur and city government can be satisfied with.
The algorithms Baratchi designs, she wants to make them better and better. They need to start recognizing patterns, taking into account all relationships between time and space. 'An important geographical law is that everything is related to everything, but things that are close together more than others. For example, the temperature in a city is often closer to the temperature in a nearby other city than to the temperature in a city farther away. I try to make my algorithms smart enough to understand that and use that information.
Baratchi also wants its algorithms to be able to use data from different sources, collected with different techniques and of different quality. The algorithm must be smart enough to adapt to the data it is presented with. For example, Baratchi is supervising a doctoral student researching conservation in the Oostvaardersplassen. 'Here we have to generate data ourselves, because satellite data on vegetation development are not enough. We have therefore hung cameras to monitor the behavior of the large grazers in the area. We hope to contribute to optimal decisions about desired population size and interventions.'
Although data-driven policy is not yet booming, as Baratchi would like, it already offers answers to real questions. For example, about the effectiveness of appropriate education for vulnerable children. 'We have just started the Centre for BOLD Cities' (Big, Open and Linked Data, ed.) within Leiden-Delft-Erasmus. 'We are finding out whether children with autism in the concept of appropriate education come into their own in an ordinary school. Children wear sensors on their clothes at recess to measure whether they are close to other children or somewhat secluded.'
Where and how long children are in close proximity to each other provides an indication of their social interaction. Also, that information can provide suggestions for what you can improve about the schoolyard, for example, to create a more inclusive environment. With questionnaires alone, you would never get this information to surface correctly and in detail.
Students in the Computer Sciences program with specializations such as Data Science and Artificial Intelligence learn from Baratchi, first, what urban problems could be solved with data. They also learn what data they need to do so and to think creatively about where to get that data. Finally, they learn to design algorithms that can extract the relevant information from all that data.
This college year, the teacher thought her students might want to look into corona, a fine example of a spatio-temporal process. 'They started thinking about how to model the epidemic to reduce the uncertainty about it. They put the policies of different countries side by side to learn to predict the outcome of each policy choice based on outcomes of different measures.' The students started in February, so they couldn't do very much before the university closed, Baratchi puts into perspective. But that her field of research has great potential is obvious.
This article can also be found in the files Coronavirus and Big Data
Source: Leiden University
