cyvy Research Project
When it comes to machine learning, it is all about the data. Machines that are fed large data sets learn to independently discover regularities in the data – just as the brain draws conclusions from observing the environment over and over again. The availability of large data sets has played a major role in recent breakthroughs in machine learning. In recent years, data sets have enabled or improved language processing, speech recognition, or computer vision, to name just a few examples. Some researchers even argue that data is more important than new algorithms.
However, this is different for machines and engineering systems such as autonomous robots or self-driving cars. In the domain of physical systems, data sets have not yet led to comparable breakthroughs.
Sebastian Trimpe and his “Intelligent Control Systems” research group are focusing on the question of whether rich and high-quality data sets collected on machines (for example, data collected on an engine) can be similarly beneficial for learning in the context of physical machines. The team will collect diverse data sets on a variety of physical machines in collaboration with Cyber Valley industry partners and empirically investigate whether such data is useful for a learning task on a different machine.
In addition, they will address conceptual and fundamental research questions such as: When is data sufficiently informative for a dynamic learning problem? What data can be meaningfully transferred from one machine to another? And how can available data sets be used for modeling and control on new machines?
The developed machine dynamics data sets and benchmarks, as well as all scientific results are planned to be made publicly available to the research community.