The ongoing digitalization and technicization of our everyday lives is a challenge that scientists at TU Wien are facing together with researchers from IARAI. While the cooperation is anchored on TU Wien's side at the Department of High Performance Computing, the Institute of Advanced Research in Artificial Intelligence (IARAI), with locations in Vienna, Linz and Zurich, complements its expertise in artificial intelligence (AI) to the cooperation. By signing the contract in April, the two partners announced their cooperation.
Long-term cooperation planned
At first glance, the partners seem to be quite different: While IARAI has extensive background knowledge in the field of machine learning, the TU Wien has extensive experience in operating such systems and possesses technical know-how. However, it is exactly these different starting points and working environments that can lead to an exciting and successful collaboration, the researchers agree.
Once the infrastructure is in place, the research team will conduct its first experiments, including high-performance networks, fast storage connections and various software stacks. In addition to the technical implementation, the inner attitude is crucial: "An important aspect in this early phase is to be open to new ideas and innovative approaches," says Dr. Michael Kopp, Founding Director at IARAI. In order to further consolidate the connection between TU Wien and IARAI, not only the most modern hardware is being established at TU Wien, but a lively exchange of technical experts from IARAI and TU Wien is being set up, with IARAI employees also working directly at the TU Wien.
The technical challenges
"During the first phase, we will share existing knowledge and benefit from each other's expertise," says Dr. David P. Kreil, Founding Director at IARAI. Furthermore, existing resources will not only be used more efficiently through the cooperation, new hardware will also be installed at TU Wien's dataLAB. This modern equipment includes the latest GPU systems from NVIDIA (A100 GPUs) combined with a high-speed network (internally 4,800 Gbit/s, externally 200 Gbit/s) as well as a high-performance storage system in the dataLAB. Although such GPU-based systems represent a major challenge for the development of highly complex systems, they are currently very popular for AI applications.
Therefore, as part of the cooperation, alternative non-GPU-based AI systems will be investigated and evaluated as part of the collaboration.
Creating AI for humans
"In the context of the collaboration, we primarily want to understand and optimize fundamental (IT) processes," says Dipl.-Ing. Bernd Logar, CIO of TU Wien. "Currently, the decisions of AIs are difficult to understand and verify - not only for users, but also for the developers themselves". This understanding is, therefore, urgently needed for the social acceptance of AIs.
Although the scientists primarily study computer systems, humans play a central role in machine learning research: "As controversial as artificial intelligence is seen and discussed, especially in connection with ethical issues, in the end it is about how to enable people to live a good and self-determined life in times of advancing digitalization," summarizes Dr. Ernst Haunschmid, Head of dataLAB, TU Wien IT Solutions.