Research Center Machine Learning

Research Areas

Resource Aware Learning

Tailoring algorithms to individual devices

Machine Learning algorithms need to be tailored to the devices they are running on. Edge computing on IoT devices requires different solutions than learning on large clusters in computer centers and quantum machine learning deals with yet another kind of architectures.

Simulation-based Learning

Integrated physics-based simulations

In many application areas, data are still scarce but may be created by means of simulations. Simulation-based learning integrates physics-based simulations, generative modeling, and resampling techniques to produce plausible representative training data as well as to improve simulation models.

Hybrid Learning

Combining knowledge and data

Conventional Machine Learning methods yield suboptimal results if training data are biased, not representative, or otherwise lacking. Hybrid learning algorithms address these issues by combining data- and knowledge driven techniques. Incorporating domain knowledge into structure or design of learning systems can provide them with common sense and leads to explainable solutions.