Machine Learning

Fraunhofer SCAI develops new intelligent methods for machine learning and adapts data analysis methods to specific use cases. SCAI picks up the knowledge available among users, structures it and incorporates it into mathematical or semantic concepts that flow into data-driven models as domain knowledge.

SCAI also is investigating the mathematical foundations of learning methods. We develop new approaches to integrate knowledge from applications and methodologies to transfer experience in a structured way. Furthermore, SCAI is working on the scalability of data analysis methods.

Projects and Application Areas

 

Projects

In numerous projects, SCAI develops new methods in the field of machine learning and artificial intelligence and applies them in many different use cases.

 

Interpretation of Patient Data

Using text-mining methods, the procedures developed by SCAI enable a fast and automated overview of existing knowledge in current medical literature.

 

Data-Driven Energy Management for Networks

SCAI offers optimization methods and tools for the analysis of sensor data in combination with modeling and simulation for the data-driven energy management of networks (electricity, gas), production facilities and energy-intensive infrastructure.

 

Digital Twins

SCAI develops new methods for context-dependent model generation and coupling.

 

Virtual Product Development

SCAI develops innovative machine learning methods to understand the complex physical and technical phenomena studied.

 

Innovative Material Design

SCAI combines and develops methods and tools of machine learning, data analysis, multiscale simulation and high-dimensional optimization in the approach of data-driven virtual material design.

 

Predictive Maintenance

The predictive maintenance of technical systems is important for many industrial applications. SCAI develops machine learning approaches to improve detection quality and robustness.

 

High Performance Computing

Machine learning methods play an important role in analyzing large amounts of data and embedded systems, where certain tasks have to be reliably solved with limited computing power.

 

Research Center for Machine Learning

The direct transfer of ML research to industry is the aim of the Fraunhofer Research Center for Machine Learning within the research cluster »Cognitive Internet Technologies«.

As an own research field, HPC develops scalable and reliable algorithms, efficient software systems, and new ­hardware solutions. It thereby covers the whole range of computer ­architectures, from huge clusters of interconnected computers to various kinds of specialized computing facilities. The latter includes accelerators like many-core processors, graphics ­processing units (GPUs), ­microcontrollers, or field ­programmable gate arrays (FPGAs) for use in machine embedded computing. The goal is always to gain optimal performance out of the given hardware, e.g. efficiency in terms of reaction time or power consumption. Furthermore, the reliability of algorithms and computed results is an essential issue.

Machine Learning (ML) stands for a revolutionary new ­paradigm in science and engineering. In ML, data is ­recognized as a driving force which leads to new knowledge and changes the way of thinking in computational sciences. For realistically-sized problems, ML and its applications require a systematic use of HPC technologies. To govern the ­complexity of these applications, algorithms, software and hardware have to be carefully adjusted.

Fraunhofer SCAI, based on its expertise in numerical ­simulation and algorithms, combines HPC and ML for breakthroughs in industrial practice. SCAI’s work is focused on the development of new and intelligent algorithms and on the adaptation of data analysis methods to the specific needs of application projects. Data is not only taken from ­measurements or sensors, but combined with and enriched by results from physical modeling or numerical simulations and general application knowledge. Such an approach is ­often called a gray box. In particular, in ­engineering and natural sciences applications, this approach offers more ­reliability in predictive analyses and enriches the potential of ML ­technology. It differs from a black box approach which takes data as it is and treats it with generic data analysis methods, thereby neglecting available specific domain knowledge and structure.

SCAI offers its broad experience in ­computational science and engineering as well as its deep knowledge of HPC systems and ­programming for algorithmic developments and machine learning applications.

This site gives an overview on ­activities in HPC and ML, presenting ­applications as well as algorithmic approaches.