Paradigm shift in applied research

AI series: How artificial intelligence advances technologies and creates new solutions for customers.

A computer that recognizes coins based on the sound they make? That would be a walk in the park for the researchers at Fraunhofer IIS today. But what about over 25 years ago? “We actually had a functioning recognition system back then; researchers used an artificial neural network, which they trained,” says Dr. Thorsten Edelhäußer, head of research planning at Fraunhofer IIS. “So machine learning was an occasional topic here from as far back as 1992. Back then, only a few ingredients were missing to give us artificial intelligence.”

Today the ingredients are there in the form of high-performance computers and huge data volumes. But that’s not all: “Over the past ten years, we’ve built up application- and customer-oriented expertise that has dug deep into the subject and explored practical problems close up.” What’s more, Edelhäußer is currently experiencing a paradigm shift in applied research at Fraunofer IIS.

For various problems that were previously unsolvable with conventional methods, AI is now stepping in – or rather, previous methods are being “enriched” and thus improved with AI. Fraunhofer IIS predominantly uses machine learning. Machine learning is a sub-field of artificial intelligence that seeks to move away from explicitly programming computers toward enabling them to learn independently from existing data.

Dr. Thorsten Edelhäußer and Ilona Hörath
© Fraunhofer IIS/Paul Pulkert

Dr. Thorsten Edelhäußer, head of research planning at Fraunhofer IIS, talks to journalist Ilona Hörath.

Deep Learning System
© Fraunhofer IIS/Franziska Köhler

Machine learning involves algorithms that are able to autonomously learn based on existing data. The algorithm identifies regularities in the data and derives analytical models. Machine learning can be realized with so-called neural networks composed of connected artificial neurons (blue dots). The neurons are arranged in an input layer, an output layer and one or more hidden layers, which are layers of neurons between the input and output layers. The figure shows machine learning on the basis of multipath problems as further explained in the article.

Edelhäußer uses an example to explain: “It’s well known that you get poor signal reception when you drive through narrow, built-up areas with tall buildings. In such circumstances, the satnav may even display the wrong street. The reason is that the signals coming down from the satellites are reflected off the walls of the buildings and arrive at the vehicle’s antenna with a variety of time delays. This distorts the displayed position. Ordinary algorithms struggle to cope with this so-called “multipath problem.” However, machine learning has opened up a new way of solving such conundrums.

As such, AI helps deepen and broaden the expert knowledge of researchers at Fraunhofer IIS. “We can find solutions for our customers that were previously inconceivable.”

Interview by Ilona Hörath.

 

 

Deep Learning System
© Fraunhofer IIS/Franziska Köhler

Machine Learning

Click on the image to enlarge

Contact

Your questions and suggestions are always welcome!

Please do not hesitate to email us.

Stay connected

The newsletter

Sign up for the the Fraunhofer IIS Magazine newsletter and get the hotlist topics in one email.

Home page

Back to home page of Fraunhofer IIS Magazine