Mahyar Tourchi Moghaddam, who is Assistant Professor at the Maersk Mc-Kinney Moller Institute at the University of Southern Denmark, is very interested in this area. Moghaddam is associated with DIREC and the project ‘Cyber-Physical Systems with Humans in the Loop’, which will explore and map out how software engineers can best design cyber-physical systems adapted to users and the context in which they are to function.

Read more in Danish.

As robots and automation solutions increase in industrial companies, plans and project descriptions grow in numbers, often ending up in a digital folder on a company PC – if such data is stored at all. In most cases the data is never to be used again.

So is the reality today, but in the future, established solutions for automation, robots and data produced during the production process shall be reused, possibly even from other companies. In that way, companies do not have to design the complete solution from scratch themselves.

This is the ambition of the ReRoPro project (Re-Use of Robotic-data in Production through search, simulation and learning) which is funded by the national research centre DIREC and led by the Mærsk Mc-Kinney Møller Institute at SDU.

– Big data is already being used in the IT-field, where it has been crucial for developments in areas such as facial or object recognition. In the robotics field, there is great potential for gathering data and experience from companies that have already adopted new technology and automation in their production, says professor Norbert Krüger from the SDU Robotics at the Faculty of Engineering.

– Today, each company stores robot data – if at all – in their own format, but such data could just as well benefit others so that we can spread the use of robots and ultimately maintain or even attract production back to Denmark, he says.

Giants onboard

SDU is behind ReRoPro, together with Aalborg University and University of Copenhagen who also are partners in the DIREC project. The project also includes two Danish industrial giants, Novo Nordisk and Rockwool, the robotics company Nordbo from the Funen robotics cluster and the Allerød based company Welltec. Odense Robotics and MADE are also partners.

The ambition is to create a structure for a database with the help of the companies. We aim to gather information from existing solutions in such a structure that makes it possible to reuse or be inspired by that data when creating their own solution, says the SDU professor.

Novo Nordisk and Rockwool are already on board, but the intention is to get even more companies involved, Norbert Krüger stresses. That’s why a conference is planned for 8 September, where he hopes many industrial companies will come forward.

– For a lot of companies, this kind of data is considered a company secret, so we need to find a way how they can learn from each other in a safe environment. At the same time, we want to know more about what companies need out in the field so that we can take that into our work, says Norbert Krüger.

Initially, the project will run until autumn. Still, the plan is to outline a plan and invite industry along – via the conference – so that funding can then be secured for a significant research project based on the initial work, which is starting now.

Read more about the conference
Read more about the project

By combining this feature with cryptographic techniques will enable banks and authorities to share sensitive personal data securely and enable them to fight fraud. This is exactly the purpose of a new project between researchers from Aarhus University, the IT University of Copenhagen and the Alexandra Institute, which is supported by DIREC – Digital Research Centre Denmark.

Read more (in Danish)

Miao works as Assistant Professor at the Department of Computer Science at Aalborg University. He is also part of the DIREC workstream Advanced and Efficient Big Data Management and Analysis. The project focuses on how we can develop new efficient prototypes that can enable the use of big data in industry. Miao focuses especially on building efficient and explainable prototypes for different tasks and data in an automated manner.

Can you tell us about your background and why you settled down in Denmark as a computer scientist?
I am interested in machine learning, automated deep learning and explainable AI. I hope that I can introduce automated deep learning and explainable AI to the Danish data science community, since research about this topic is rare.

Besides that I chose to come to Aalborg because it is a young and very active university, which provides a lot of opportunities for young researchers. I have several friends, who are working here, and they also recommend me to join their group, Center for Data-Intensive Systems (DAISY), which has an international reputation. I believe I can learn a lot here. 

I think the working environment in Denmark and Aalborg is pretty good. We have a lot of flexible time, so I can focus on my research. In addition, I think Aalborg is an environmentally-friendly city, and I really enjoy life here.

Can you tell us about your research area?
I have broad research interests in machine learning and artificial intelligence – especially automated deep learning and explainable AI. I am interested in automatic development of efficient, scalable and robust algorithms for machine learning, data mining, data management and deep learning applications with formal theoretical guarantees and explanations. I see myself working on these problems in my foreseeable research life. 

What are the scientific challenges and perspectives of your project?
Although the techniques of deep learning have been applied in different areas, such as computer vision, face recognition, medical imaging, natural language processing, data mining and data management, the design of deep learning systems is time-consuming – and it is still a black box problem to explain why the developed deep learning system is working. 

Automated deep learning is the process of building deep learning systems for different problems without human intervention. Explainable AI is to explain why the developed system is working – and it can also assist the design of the deep learning system. Automated deep learning and explainable AI are in their early-stages, and we still need to define some research problems, improve efficiency, and explain why the automated designed system works.

How can your research make a difference for companies and communities?
Automated deep learning aims to build a better deep learning system in a data-driven automated manner, so that most practitioners in deep learning can build a high-performance machine learning model without being an expert in the field of deep learning.

Automated deep learning can provide end-to-end deep learning solutions and these solutions are usually better than hand-designed deep learning systems. These automated systems can lower the threshold of deep learning and make it easy for everyone to use these techniques to solve their own problems.

Zimmermann is part of the DIREC project Verifiable and Safe AI for Autonomous Systems. The aim of the project is to develop methods and tools for safety critical systems within a variety of domains. Here, he works on understanding the foundations of correct and secure systems.

Can you tell about your research area?
Software and embedded systems are everywhere in our daily lives, from medical devices to aircrafts and the airbags in our cars. These software systems are often very complex, and it is challenging to develop correct systems. Therefore, we need verification software that can check such systems for errors.

The news is full of stories of potential vulnerabilities in software and embedded systems. Some of these vulnerabilities have been there for several years and are very hard to find. They might not be seen in daily use – only when you try to exploit a system.

It is even more pronounced when you look at distributed systems made up of several components interacting with each other. Like a website for the seat reservation system in a cinema where you click on the seat, which you want to book, while others do it at the same time. The system must be able to deal with many concurrent requests. Verification tries to automate the reasoning and automatically proves that the system is correct and safe.

How can we make these systems more secure?
Personally, I am interested in viewing this as a kind of game. I want to design a system that lives in an environment, so I understand this as a game between the system and the environment. The system wants to satisfy a certain property and the environment wants to break the system. And by that game-view you can get very strong guarantees.

It’s very hard to get complex systems correct. And if you have a safety-critical system you need those guarantees to be obtained by verification software. If you employ software that controls an airbag, then you want to be sure that it works correctly. It’s easy to miss errors – so you cannot rely on humans to check the code.

What is the potential of verification?
Verification is a very challenging task. It is challenging for a human to argue that a system is correct, and it is also hard for a computer, so unfortunately, it is not applicable universally. Verification is used for systems that are safety-critical, but even here there is a tradeoff between verification cost and development cost. 

One of our goals is to develop techniques that are easy to use in practice. We work on the foundations of verification languages and try to understand how far we can push their expressiveness before it becomes infeasible to verify something. It can take hours or days to verify something, so it is a computationally expensive task. We try to understand what is possible and try to find problems and application areas where you can solve this task faster.

Another important thing is that we need precise specification languages for verification. You cannot use natural language. The verification algorithm needs a precise specification with precise semantics, so we are developing different logics to see if they can be used by engineers to actually write specifications. If it is too complicated for the practitioner, e.g., the engineer, it will not be used. You must find the sweet spot between expressiveness and usability.

Did you know Aalborg University before you were employed?
I have had a connection to Aalborg since my PhD where I worked on a European project with partners from all over Europe including the DEIS group in Aalborg. I was in Aalborg a few times during my PhD and knew people here. Aalborg is central in Europe when it comes to verification and design of systems. There are many collaborators and there is a good connection to the industry compared to other places. It is a very good location.