The Max Planck ETH Center for Learning Systems has just been renewed for another five years. Co-director Prof. Thomas Hofmann explains the role he sees for the center in Europe.

Prof. Thomas Hofmann
Prof. Thomas Hofmann

 

The Max Planck ETH Center for Learning Systems (CLS) is the first joint research centre between ETH Zurich and the German Max Planck Society (MPG). In May 2020, it celebrated its fifth anniversary and was extended for another five years. In this interview, the centre’s co-director Professor Thomas Hofmann takes stock, looks to the future and explains why the topic of artificial intelligence requires thinking beyond national borders.

Professor Hofmann, the Max Planck ETH Center for Learning Systems (CLS) has completed its first five years. Are you satisfied with how the centre has developed during this initial phase?
Yes, the centre is very active and arouses a lot of interest within the scientific community and the general public. Machine learning and artificial intelligence are widely discussed today. When the CLS was founded five years ago, the general population barely knew about the potential of these technologies. In this respect, the centre was a few years ahead of its time.

Winfried Kretschmann, Minister-President of Baden-Württemberg, mentioned the CLS in a working meeting with Federal Councillor Guy Parmelin. He stated that even closer cooperation in research would be of great benefit to both sides. How do ETH Zurich and the Max Planck Society benefit from the centre?
The CLS gives us the opportunity to attract international talents. Working at the centre is appealing to doctoral students and postdocs because the CLS offers more freedom and more critical mass than a research group. The exchange of personnel is also very valuable. Even within ETH Zurich, the CLS has brought about collaboration between researchers from different departments who would have rarely met at ETH. And finally, the CLS increases the visibility and reputation of our institutions. Both ETH Zurich and the Max Planck Institute for Intelligent Systems already enjoy an excellent international standing. By working together, we can strengthen it further, especially in the competition for talent, where we are up against industry giants like Google and Amazon.

 

"Even within ETH Zurich, the CLS has brought about collaborations between researchers who would have rarely met at ETH."Prof. Thomas Hofmann

Together with Professor Bernhard Schölkopf, you are co-director of the CLS. What is your role?
After founding the CLS, it was necessary to define the operational framework, advance the scientific agenda, promote cooperation and build a community of researchers. My task was also to integrate the diverse environment of ETH Zurich into this cooperation. While the Max Planck Institute for Intelligent Systems is a single organisational unit, various departments at ETH Zurich are involved, including the Departments of Computer Science, Information Technology and Electrical Engineering, Mechanical Engineering and Mathematics. Finally, we defined the strategic direction of the centre in Europe and sought to attract young talents for our doctoral program. For this purpose, we developed a centralised selection process with strict quality requirements.

Has the centre already contributed to research?
We are not working on a single major research project, but are pursuing a wide range of research questions. Since our goal is to connect ETH Zurich and the Max Planck Institute for Intelligent Systems, one of the criteria we use to measure our success is the number of joint high-profile scientific publications. Numerous publications have already been produced in areas ranging from robotics and computer vision to causality and theoretical fundamental research, which I am working on, too.

"Our fundamental research does not only bring benefits in the distant future – we answer pressing questions that arise today."Prof. Thomas Hofmann

With topics like robotics, the potential benefits are obvious. What are the benefits of theoretical fundamental research?
One of the big problems in machine learning is that we can train these gigantic models with huge amounts of data for various tasks, but we do not sufficiently understand how and why they work. For example, models often do not map the causal relationships in the world, but make predictions based on observational data and the statistical correlations and dependencies contained therein. However, causal relationships need to be taken into account to make our models more transferable, reliable and transparent.

At the same time, these theoretical questions are very relevant to practice. Our fundamental research does not only bring benefits in the distant future – we answer pressing questions that arise today. There is also a certain division of labour between academic research and industry. We cannot compete with large companies that use these models for a variety of commercial applications, but we can lay the foundations for better models.

What will the next five years bring?
We want to continue our research, doctoral programme and public communications. We also plan to enable Master’s students to do research at the centre. At the same time, our aim is to make the CLS a nucleus for a European network of researchers, even beyond the next five years. Today, the field of artificial intelligence is confronted with high expectations from politicians and the public. Many countries have launched national research initiatives. Research institutes in Europe must become organised and join forces, which takes time. Since the CLS started very early, we are already up and running and can now grow organically. Other initiatives also benefit from our experience, such as ETH Zurich’s own Center for Artificial Intelligence, which is currently being established, or the European Laboratory for Learning and Intelligent Systems (ELLIS), which networks scientists throughout Europe.

"If European countries do not think beyond their borders, they will have little success."Prof. Thomas Hofmann

Wouldn’t it be better to concentrate all forces in one large institution from the outset instead of founding so many different centres and networks?
It is indeed our “secret plan” to merge the activities of these institutions, for example in a Europe-wide ELLIS institute, which could also include the CLS and the ETH Center for Artificial Intelligence. The greatest lasting success for me would be if the centre were to be absorbed into something bigger. However, it is difficult to create a large institute out of nothing. That is why we are slowly approaching the issue through various smaller initiatives.

Does Switzerland need a national AI initiative?
Instead of founding a national centre, we should combine existing centres of excellence in AI and use them to contribute to a pan-European network. Major global players such as the USA and China are also active in this field. If European countries do not think beyond their borders, they will have little success. If, on the other hand, Swiss expertise is placed in a European context, Switzerland can not only contribute a great deal, but also gain global visibility and considerable influence in Europe.

Max Planck ETH Center for Learning Systems

The Max Planck ETH Center for Learning Systems (CLS) is the result of a collaboration between ETH Zurich and the Max Planck Institute for Intelligent Systems in Tübingen. It is the first joint centre between ETH Zurich and the Max Planck Society. Since it was founded in 2015, the CLS has been aiming to network researchers in the field of artificial intelligence and learning systems, promote fundamental research and train doctoral students.

The centre is managed by the co-directors Thomas Hofmann (ETH Zurich) and Bernhard Schölkopf (Max Planck Institute for Intelligent Systems). Its members include several faculty members from the Department of Computer Science, including Joachim Buhmann, Gunnar Rätsch and Andreas Krause, as well as numerous other professors, postdocs and doctoral students from various departments at ETH Zurich.

external page Website

JavaScript has been disabled in your browser