Over the past few months, the Fraunhofer EAS division has been working intensively on digitalization and artificial intelligence in Saxony as part of the KIKiS project. Can you give a brief outline of the main findings and tell us what conclusions you would personally draw from the work?

We teamed up with the TU Dresden on this project and our work involved us examining the status quo and the future prospects of artificial intelligence solutions in Saxony, specifically how companies that offer AI solutions can gauge what actual demand there is for AI solutions as well as what potential AI offers, what challenges it presents and what conditions are needed to be able to develop successful AI solutions in the first place. To help us answer these questions we conducted a series of interviews with experts from research institutes and private enterprises focusing on a variety of areas, from machine learning to artificial neural networks.

Many of the experts interviewed see Saxony as having a good basis for the development and advancement of AI expertise, although they also see shortcomings, especially in knowledge and technology transfer, education and teaching, access to analyzable data sets, and employee qualification and training. For me, this means that in future everyone involved will have to work toward making knowledge exchange and transfer between research institutes, educational establishments, and companies more efficient and, more importantly, ensuring that this is a reciprocal process. This will be the only way that Saxony can shift up a gear on the competition side and gain an edge on other AI solutions on different levels, over and above the success of individual solutions. We as a research institute would like to be able to play our part in helping to achieve this.

What know-how can Fraunhofer IIS/EAS bring to collaborations with businesses and other research institutes?

For one, we are working on development methodologies for AI-based systems. One of the areas we look at in this context is how AI-based systems that are aimed at dedicated applications can be designed such that they meet all the relevant requirements, whether in terms of efficiency, reliability or costs.

We also employ various AI data analysis or data modelling methods that could be used, for instance, to monitor production machinery as part of a predictive maintenance solution. We go beyond mere data analysis, however, instead processing data detected by sensors with the help of physical models and expertise. This way, we ensure that our solution can deliver reliable results in practice. The biggest hurdle is how to process the data inline, which is needed to predict potential equipment failure in real time. This means we need fast data processors, optimized algorithms, and powerful hardware, for example.   

Can you tell us about one of the division’s ongoing AI projects – one that particularly stands out?

One of the main aims of our research is to open up our findings to small and medium-sized enterprises – and this of course applies to artificial intelligence techniques as well. When it comes to artificial intelligence, many enterprises feel stumped. They are often unable to properly gauge the potential or risks, the costs and economic benefits. What we are seeking to do is build a bridge between standard AI applications and the use of corresponding algorithms in small and medium-sized manufacturing enterprises. To do so, we’ll be drawing on the latest developments in machine learning, deep learning, and explainable AI with a view to adapting them for use in specific applications in these enterprises. To help us to do this, we’ll be using a modular approach based on tools for data acquisition, visualization, and decision-making based on AI. Test labs where these technological modules can be used are intended to ensure that these firms can get on board the AI train with as few hiccups as possible. At the moment, we’re in the preparation phase of the project.

And one last question: what does the future hold for IIS/EAS with regard to the development of AI?

The current focus of many different AI research projects is memory and storage – which are instrumental in enabling AI. Conventional computer architectures are simply not equipped for the huge quantities of data and complex algorithms. This is why many in the AI research community are working on computing of the future. We hope to be able to contribute in the fields of quantum communication and neuromorphic architectures – innovative data processing and transfer concepts that call for the development of new semiconductor and electronics architecture, for example. However, this will require entirely new system design approaches and dedicated high-performance algorithms.