bs1.png bs3.png
Navigation
blank.gif blank.gif
About us
The IMPReSS project is a 30-months EU-Brazil cooperative research project started in 2013.

The project is partly funded by the European Commission under the 7th Framework Programme in the area of EU-Brazil Research and Development cooperation under Grant Agreement no. 614100

The Brazilian funding is provided by CNPq Conselho Nacional de Desenvolvimento Científico e Tecnológico

 

 Impressum   Privacy
Share this
Bookmark and Share
Facebook
Our Newsletters
Sign In
blank.gif
Enter Username

Password



Forgot Password?
blank.gif
Crossing over the Atlantic: Tales from a Fraunhofer FIT Research Associate’s experiences at the Federal University of Pernambuco in Brazil
blank.gifCreated February 11 2016 12:48:15 pibe.gif Posted by blank.gif
blank.gifÁngel spent a month working alongside a fellow IMPReSS researcher, Lucas Lira Gomes, at the Networking and Telecommunications Research Group (GPRT) Lab facilities at the Universidade Federal de Pernambuco (UFPE). As a team, Ángel brought his knowledge in Complex-Event Processing (CEP) and Lucas his knowledge of Machine Learning. By joining forces, they designed an experiment to prove the feasibility of running a system at the gateway level.

Ángel explains: “Machine Learning (ML) is a series of advanced algorithms for pattern discovery by empiric input. Complex-Event Processing, also known as stream processing, is a technique developed to process the data as it comes, allowing to process non persistent data. Combing this techniques with other IoT communications protocols, i.e. MQTT, allowed us to perform automatic learning in the cloud as well as at the edge of the network (e.g. at gateway level). This technique exploits the computational power spread in the network.”

“Using the IMPReSS SDP, an array of sensors was deployed in the lab such as consumption sensors and motion sensors. To process this Data in real-time, the IoT Service agent was put in place“, Ángel explains.

The service, named IoT Learning Agent, uses CEP queries to pre-process data in real-time in order to be later on fed-in into the increasing learning. The CEP engine provides the tools needed for pre-processing the data efficiently and tools for real-time data management to deploy rules after the learning phase. ML provides the pattern recognition that allows you deploy more complex rules otherwise not possible. The learning algorithm Stochastic Gradient Decent implemented in Weka was used for the experiment.

“The work proved to be harder than expected; hardware constrains, poor quality data and loss of some of the sensor datasets were some of the problems we encountered”, Ángel tells.

Back in Germany, Ángel continues to put his final touches on IMPReSS as the project is rapidly reaching its end. He is also busy writing an article on the early results from the experiment in Recife. On a personal level, the experience of living and working in Brazil for a month has been priceless.

You can read more about Ángel’s experiences in Brazil in the upcoming IMPReSS newsletter.
blank.gif
blank.gifblank.gif