ReBAR
Artificial Intelligence for Everyone
The increasing spread of machine learning methods opens up enormous development potential in a wide range of application areas - such as engineering, transport and medicine. There is now a wide range of methods available with a bewildering variety of implementations. The project partners in Reducing Barriers for AI in (applied) Research – ReBAR have set themselves the goal of simplifying access to machine learning methods. We are creating a modular platform that works independently of the specific technical implementation. In addition to the Institute for AI Safety and Security, the DLR Institutes of Structures and Design, Robotics and Mechanics, Software Technology, Materials Research, Vehicle Concepts, Maintenance and Modification and System Architectures in Aeronautics are involved in order to not only implement the core functions by June 2024, but also to create the core of a user community at DLR.
Development of an accessible machine learning platform for a broad spectrum of applications
The idea behind ReBAR is simple: the entire work process from the raw data to the evaluation and verification of the results should take place in a single environment. Depending on the specific data sets and the research question, suitable methods are loaded as modules in the environment. The individual modules are linked with standardised interfaces, allowing the methodological portfolio to be easily expanded. Thanks to the containerisation of the environment, ReBAR can be run on a wide variety of devices.
This approach offers decisive advantages for users. It allows users to concentrate fully on content-related issues without having to familiarise themselves in depth with the technical intricacies of implementation, supported by comprehensive documentation. The comparison of different methods and the assessment of the reliability of the respective models can be realised within the same environment by selecting different modules. This reduces the entry barrier to the development of AI applications and machine learning can be rolled out quickly and easily in applied projects. Developers also benefit from the standardised definition of the modules: New methods that are developed with interfaces for the ReBAR system can be immediately integrated into workflows and are quickly disseminated in the user community.
Contribution Institute for AI Safety and Security
The contribution of our institute comprises the two focal points of architecture definition and verification & visualisation. This means that our scientists are directly involved in setting up the modular structure and defining the interfaces and scope of the developed modules. In doing so, they pay particular attention to user requirements, which are derived from application examples from various institutes. We focus on analysing the results and presenting them in the form of a dashboard. This allows the models and results to be interpreted and enables further options for action to be evaluated. In particular, we present metrics and key figures that allow a comparison of different models and methods.
Five applications will be implemented and made available as examples during the course of the project. The applications include the investigation of flight data, the characterisation of materials and identification of anomalies, as well as the preparation of simulations for engines and configurations for new aircraft concepts. The ReBAR environment will be extended to other applications outside the project. For example, the integration of quantum AIs and benchmarks from the Quant²AI project is planned in a follow-up activity.