The project deals with the dynamic adaptability of the inference speed of neural networks. In future, it should be possible to react to changes in boundary conditions at runtime.

Project group: Prof. Dr. Wolfgang Karl, Roman Lehmann
Contact: Roman Lehmann
 

Neural networks are used in numerous areas of science and industry. In conventional networks, the individual layers are passed through one after the other in order to arrive at an output with the last layer. However, in order to be able to react to dynamic boundary conditions, this fixed structure needs to be loosened. In future, it should be possible to reach an output at any time (anytime). Dynamic adaptation to a production cycle is conceivable here in a real-time scenario, for example.