The ability to learn from experience is usually thought of as a specifically human (or more generally biological) attribute. But modern-day machines are also capable of automatic learning. Computers no longer need to be explicitly programmed for specific tasks. They can now be trained in a data-driven manner to tackle broad classes of problems: The greater the range of examples they have been exposed to, the more reliably they can discern regularities and patterns in the data. In this way, they learn to relate data points to each other, draw conclusions and make predictions. And at the point where such algorithms have acquired the capacity to generate appropriate actions from observational data in complex and dynamic systems, one has crossed the boundary that separates machine learning from artificial intelligence. Already, machine learning algorithms have become an integral part of speech-processing devices, and find use in face recognition and image processing programs. They also provide the basis for the development of diagnostic tools in medicine and autonomous processing systems in areas such as logistics and industrial manufacturing. The commercial potential of such applications is regarded as enormous, which explains why the term ‘artificial intelligence (AI)’ has so rapidly established itself in public discourse.
Now researchers based at LMU and the Technical University of Munich (TUM) have joined forces to set up the Munich Center for Machine Learning (MCML). The new Center will accommodate 15 teams specializing in the fields of data science, informatics and statistics. “The Center is designed not only to stimulate basic research in the field of machine learning, but also to make users in science and business aware of its potential,” says Professor Thomas Seidl, who holds a Chair in Database Systems and Data Mining at LMU, and Coordinator of the MCML. Thus, issues relating to practice-oriented research and the development of novel methodologies will make up a large part of the MCML’s mission. The new network, which also includes other research institutions and commercial entities, will therefore seek ways of translating innovative concepts from the field of machine learning into the commercial sphere. The Federal Ministry for Education and Research (BMBF) is providing 7.5 million euros in funding over four years for the MCML. The BMBF will provide financial backing for a total of four such Centers in Machine Learning. Project management lies in the hands of the German Aerospace Center (DLR).
The MCML will focus on a specific set of applications – settings in which its scientists believe that machine learning concepts can be profitably developed, implemented and evaluated. The areas selected are Industry 4.0, mobility, healthcare and the biosciences. The idea is to develop methods for use with specific types of data, such as graphs, networks, time series and sensor data. “The range of socially relevant applications for new methods of machine learning is huge,” says Professor Daniel Cremers, who holds the Chair of Computer Vision & Artificial Intelligence at the TUM, and is Joint Coordinator of the MCML.
“Automated modelling, open-source software and explainability are key components for successful transfer of machine learning methods into everyday practice. Germany also must ensure a sufficient training in statistics, data analysis and the use of algorithms for its citizens. These are precisely the goals that the MCML is intended to achieve,” says Bernd Bischl, Professor of Computational Statistics at LMU and Joint Coordinator of the MCML. Thy is why the Center will not only make the newly developed methods available for use in the applied sciences and in commercial processes, but will also apply them in training students to attack real-life problems and in the context of a carefully structured teaching concept targeted specifically at practitioners in the commercial sector.