As handheld devices and data-gathering objects become more ubiquitous, there has been an increasing desire to tap into that collective data and direct it toward new kinds of insights. Such platforms can spot potholes by picking up accelerator data on a road, for example, or microphones can be used to determine noise levels in a neighborhood.

It’s called “crowdsensing,” and it holds enormous promise for collective meaningful, oftentimes live, data. But collecting this data can have an impact on users, and finding a balance of cost and benefits can be tricky. That’s where Julia Buwaya’s research comes in.

Her project Competing Crowds explores mobile crowdsensing and the distribution and balancing of tasks and loads. It’s a subproject of the Swiss National Science Foundation synergia project “SwissSenseSynergy” of Uni Geneva, Uni Bern, SUPSI Lugano, and Chalmers Institute of Technology in Sweden. She currently works at the “Centre Universitaire d’Informatique“, an interdisciplinary center for informatics collaborating in research from citizen science to robots, with a global list of partners including Bay Area institutions such as Stanford and UC Berkeley.

Buwaya is a full-time research associate and PhD student at the Theoretical Computer Science and Sensor Lab at the University of Geneva. She has a degree in mathematics from the Technical University of Berlin, and is a former scholar of the short program course on modeling and simulation of transportation networks at MIT.

With that pedigree, it’s no surprise that she was among the three “Digital Fellows” selected in 2017 to represent the University of Geneva at swissnex San Francisco and in the Bay area. We caught up with her during her visit to Swissnex San Francisco.

What do you do?

I see myself as a researcher on the theoretical fundamentals of interaction, interplay, and human strategies. I deal with crowds – or rather, with the concept of mobile crowdsensing.

We can use crowdsensed information in a variety of ways. Right now, it’s already being used for weather monitoring in rural areas of East Africa, and in participatory citizen sensing systems for sharing information on water conditions and flooding in Vicenza, Italy and Doncaster, UK.

In mobile crowdsensing, the allocation of tasks – called load balancing – among many personal mobile devices is a fundamental and non-trivial issue. We don’t want to sacrifice a user’s costs, experience or quality of service in the collection of this data. That means tasks should be allocated so that short- and long-term system performance works for everyone with respect to costs, quality of results, user satisfaction and further application-specific metrics.

How we do that is an important question. I’m at swissnex San Francisco with my project Competing Crowds. I’m testing task-allocation algorithms based on a game theory model. I’ll also be collaborating with different research groups in the Bay Area from the field of crowdsourcing, sensing, algorithms, and game theory.

In my model, you have crowd participants and those requesting their data. Both are what I call “selfish strategic agents” with a reason to be involved in the sensing application. Basically, each one is trying to maximize what they get, with respect to the knowledge available to them. First, I’m going to test a model where participants don’t receive a monetary reward, and then, in the second phase, they start competing for a price. We’ll see which one is more efficient, and if theoretical models hold in practice. The Bay Area is a great place to do this.

I have help from the researchers and developers of a European project called Noise-Planet. They started an open-source Android app that collects anonymous noise levels (not recording conversations – privacy matters!). I’m extending the app to fit San Francisco and balanced task distribution.

The Geneva University Hospitals, in collaboration with EPFL, are already creating a noise level map of Geneva. Exposure to noise can be harmful, and the data will be used for further research in this area. We want to do this for San Francisco – but in a different way, integrating newest trends on distributing and balancing tasks and loads in mobile crowdsensing systems.

Why do you do it?

In these times of big data, I’m not only interested to see how much information and data can be collected, but how it’s collected, from a theoretical perspective: how do people and data collectors interact, what are incentives for participation, what are conscious or subconscious strategies of the agents involved? What is balance in a crowd? Equal share of work vs. everyone as they can? How do we make sure everyone is telling the truth about data, availability, ability?

One could interpret “Load Balancing in Crowds” as the counterpart to big data analytics. In place of collecting massive data and figuring out what they mean, we try to figure out what we need and how we can get it in an efficient and load minimizing way.

The results for mobile crowdsensing may also be transferable to other topics related to data generation and citizen science. You don’t have to register for a crowdsourcing application to generate data for a third person. Today, anyone is part of a crowd collecting data – willingly or unwillingly. It would be great to provide more general guidance on how to collect data from citizens in an efficient but balanced manner.

What’s NEXT? 

Everything is so overloaded. I think load balancing may build more confidence and improve trends on voluntarily participating. Fairness and balance are important. A major challenge is definitely that while minimizing loads, we want to be sure that we are not missing out on interesting information. Could we even increase the value of data if we assure a balanced distribution of loads? We’ll find out!

It’s all about balance. My team and I want to build awareness on the topic and an open source framework for load balancing algorithms in mobile crowdsensing systems. Its abstract and complex, but it’s possible!

Furthermore, in the more general field of crowdsourcing, the concept of introducing competition to improve performance already exists, e.g. on Amazon Mechanical Turk or LEGO ideas. With respect to sensing tasks and mobile phones, the concept of competition for some kind of reward is quite new.

Major advantages and challenges arise from the mobility of the users and the fact that in mobile crowdsensing, you very frequently need a cumulative product from the crowd. So you need competition, but competition that promotes collaboration of the crowd in some sense.