Jensen Huang, founder and CEO of NVIDIA, a US-based computer company, presented the new AI system in his keynote speech at the GPU Technology Conference Europe (GTC) in Munich yesterday and referred to the outstanding work of DFKI in the field of satellite image analysis.
Prof. Dr. Andreas Dengel, head of the research area Smart Data & Knowledge Services at DFKI in Kaiserslautern and coordinator of the Deep Learning Competence Centre, is pleased about the growth in the server room: "With the extension of our Machine Learning computer centre by a DGX-2, we are consolidating the position of DFKI as No. 1 in the application-oriented use of Deep Learning for industrial use and creating an infrastructure for research in the field of deep learning that is unique in Europe".
One of the outstanding research projects to be further deepened with the new hardware is the analysis of satellite images for the recognition and recording of the effects of natural disasters, from which emergency and rescue forces are supported with time-critical information.
A current key topic for the development of learning and autonomous systems (and another subject of DFKI research) is the decoding of the processing paths of deep neural networks. Novel DFKI procedures are in the process of shedding light for the first time on the processes of the "black box" of deep learning and help to make their decision-making processes more comprehensible.
In addition, the new computing possibilities are intended to open up further promising potentials.
NVIDIA describes the DGX-2 as the most powerful AI system in the world designed for the most complex AI challenges, with 2 petaFLOPS of performance delivered in a single node. The supercomputer integrates 16 NVIDIA Tesla V100 Tensor Core GPUs connected via NVIDIA NVSwitch - an AI network fabric that delivers 2.5TB per second of throughput. Its revolutionary architecture enables the acceleration of new AI model types that could not previously be trained. Thanks to DGX-2, the complexity and size of neural network models are no longer limited by the boundaries of conventional architectures.
The expansion of the DFKI's machine learning infrastructure is being funded by the state of Rhineland-Palatinate within the framework of a joint priority for the expansion of deep learning research in Rhineland-Pfalz.
