Unbiased Presentation Coach
Great public speakers are made, not born. Quantle app is a fair and honest presentation coach. It estimates the number of syllables and words you say, computes the length of pauses you make, estimates your pitch and laudness. And what's most important, it protects your privacy and doesn't share any part of your talk with cloud services as Alexa and Google Assistant do! Yes, you can use Quantle even if your phone is in the flight mode (which is more preferable anyway to minimize interference with other equipment while speaking ...)
Calibrating Noisy Mobile Sensors (or Multi-hop Linear Regression)
Today, calibrating a network of static sensors is well understood. Each sensor has to be calibrated separately. Parallel calibration of sensors is only possible if a value of the signal of interest is the same at both sensor locations. Calibrating a network of mobile sensors presents new calibration opportunities: when two sensors rendezvous, they share the same location in time and in space, and are thus exposed to the same signal value. We use sensor measurements at rendezvous to improve sensor calibration and to detect transient and permanent sensor faults.
- EWSN'12 and
RealWSN'13 papers on calibrating low-cost mobile noisy sensors by leveraging their spontaneous rendezvous.
- Our IPSN'15 paper proves that Geometric Mean Regression (GMR) used to calibrate gas sensors over multiple hops yields very low error accumulation.
- Journal paper describing SCAN, a generalization of GMR to arrays of cross-sensitive sensors, published by ACM IMWUT'17.
Constructing High-resolution Air Quality Maps
I have always been curious about the data and the dependencies, evidences and anomalies
they are hiding. In the OpenSense project, our work focused on
pushing the limits of today's air quality maps by extending the conventional network of static
measurement stations with ten mobile stations. Yearly, we obtain tens of millions of pollution
measurements from all over the city. The main challenge is to show that despite using low-cost (less precise,
low-resolution, less stable, noisy) gas sensors, we are able to construct high-resolution
air quality maps of Zurich. We explored various dependencies between measured pollutant concentrations
and many freely-available land-use datasets for Zurich such as traffic distribution, elevation, population density, etc..
A land-use regression model is now applied to construct high-resolution air quality maps
for a desired temporal resolution. Spatial resolution is as good as the resolution of available
land-used data. This work received much attention in the research community (including engineers,
environmental scientists, and health specialists) and local media.
- Our PerCom'14 paper on pushing the limits of air-quality models got the best paper award and an extended version was published in the PMC journal'15.
- Seasonal maps for 2014 are available on the OpenSense Zurich deployment page.
- An iPhone app, developed by David Hasenfratz, computes a healthier route between any two locations using the constructed pollution maps.
Analysis of Political Opinions and Detecting Election Fraud
Even though I am not actively following and participating in politics I have developed a very strong interest in the way elections are held and how they can be influenced. Main driver behind this are recent developments in my home country Ukraine which made me wonder how election data can be used to predict the fairness and accuracy of elections according to democratic principles. Around five years ago, I started learning and implementing forensic indicators of election fraud and testing them on publicly available Ukrainian, Russian, Swiss and German datasets. This work has been done in my private time and is not in relation to my activities at any research institution, hence no official publications are available. Amongst many sources of information I can recommend the following book for interested readers:
Detecting Wireless Network Boundaries
Consider a network of communicating devices deployed somewhere in an unknown or unobserved environment. Given solely a network's communication graph, how can one determine holes in the deployment and the network's outer boundary? It turns out that even suggesting a meaningful mathematically sound definition of network boundaries is challenging. Our contribution to boundary recognition literature is an accurate boundary definition and an algorithm which determines provably inner nodes of the network and proclaims all other nodes to be boundary nodes. Compared to state-of-the-art, the developed approach is scalable, range-free, and does not require high network density to compute a reasonable network boundary.
Data Routing in Low-power Wireless Nets
The first research topic I started working on as a Ph.D. student was energy-efficient routing in sensor networks. The proposed optimization leveraged the observation that a routing path is symmetric from the data yield prospective, but asymmetric from the energy consumption prospective. The latter implies that if a packet has to be lost, it is better to loose it as early as possible. This statement is somewhat counter-intuitive when thinking about getting the packets through the network, but makes more sense when optimizing energy-efficiency defined as the price for data delivery.
As a postdoc at ETH Zurich, I had the pleasure of working together with Federico Ferrari and Marco Zimmerling on a communication primitive called Glossy, which proposes a radically different approach to routing in wireless networks. Glossy is fast, energy-efficient and achieves remarkable data yields. It laid down the ground for a communication protocol stack LWB and further concepts on top of it that provide low-level guarantees. This work is a huge step forward for numerous control applications that wish to leverage wireless communication.
- An IPSN'11 paper on efficient network flooding and time synchronization with Glossy.
- My first paper on energy-efficient routing metric presented at EWSN'06.